Methods for analyzing internal defects in building walls by integrating infrared thermal imaging and acoustic data

Abstract

This invention provides a method for analyzing internal defects in building walls by fusing infrared thermographic and acoustic data. Relating to the technical field of data processing, the method includes: acquiring infrared thermographic sequences and acoustic response sequences and performing spatial registration to form a coupled observation data set; extracting temperature recovery delay and acoustic propagation disturbance characteristics to construct a defect response feature set; achieving thermoacoustic information fusion characterization through a coupled discriminant function; obtaining a stable defect response distribution field by combining spatial continuity constraints and stage consistency correction; and finally performing threshold mapping and region extraction based on this distribution field to output the wall internal defect identification result. This invention can improve the identification accuracy of internal defects in building walls under low-temperature cycling environments.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data. Background Technology

[0002] Existing methods for detecting internal defects in building walls mainly include non-contact detection methods based on infrared thermography and acoustic detection methods based on ultrasound or shock elastic waves. Infrared thermography identifies internal defects such as hollowness, delamination, and cracks by collecting temperature distribution data on the wall surface and analyzing temperature field anomalies. Acoustic detection methods determine structural continuity changes by analyzing the propagation time, attenuation characteristics, and reflection features of sound waves within the wall. In practical engineering applications, especially in cold chain storage or low-temperature engineering construction scenarios, the walls are subjected to alternating freezing and thawing environments, leading to complex changes in the internal material structure and thermophysical properties. Therefore, infrared thermography and acoustic detection methods are increasingly being combined. By simultaneously acquiring temperature field and sound propagation information within the same detection area, multi-source joint analysis of internal wall defects can be achieved, improving the reliability of the detection results.

[0003] However, in low-temperature cycling environments, existing technologies typically lack the construction of a unified characterization mechanism between infrared thermography and acoustic data, fail to perform collaborative fusion analysis of the two types of physical quantities within the same spatial and temporal framework, and lack systematic constraints on spatial continuity and multi-stage response consistency. Consequently, when faced with complex defects such as hidden cracks, insulation layer delamination, and moisture migration caused by freeze-thaw cycles, it is difficult to simultaneously ensure detection sensitivity and discrimination stability, resulting in detection results that are easily affected by environmental disturbances and insufficient defect identification accuracy. Summary of the Invention

[0004] This invention provides a method for analyzing internal defects in building walls by integrating infrared thermography and acoustic wave data, which can improve the accuracy of identifying internal defects in building walls under low-temperature cycling environments.

[0005] A first aspect of the present invention provides a method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data, the method comprising: The infrared thermal image sequence and acoustic response sequence corresponding to the target wall are acquired and spatial registration processing is performed to form a coupled observation data set. Based on the aforementioned coupled observation data set, temperature recovery delay characterization and acoustic propagation disturbance characterization are constructed to form a defect response feature set; A coupled discriminant function is constructed for the set of defect response features to achieve fusion characterization of infrared thermography and acoustic response; Based on the coupling discriminant function, spatial continuity constraint processing and stage consistency correction processing are performed to obtain a stable defect response distribution field; Based on the stable defect response distribution field, threshold mapping and region extraction processing are performed to generate a set of defect regions and output the internal defect identification results.

[0006] A second aspect of the invention provides an apparatus for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data. The apparatus is used to perform the method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data as described above. The apparatus includes an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire the infrared thermal image sequence and acoustic response sequence corresponding to the target wall and perform spatial registration processing to form a coupled observation data set. The processing module is used to construct a temperature recovery delay characterization quantity and a sound propagation disturbance characterization quantity based on the coupled observation data set to form a defect response feature set. The processing module is used to construct a coupled discriminant function for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response; The processing module is used to perform spatial continuity constraint processing and stage consistency correction processing according to the coupling discriminant function to obtain a stable defect response distribution field; The output module is used to perform threshold mapping and region extraction processing based on the stable defect response distribution field, generate a set of defect regions, and output the internal defect identification results.

[0007] In a third aspect of the invention, an electronic device is provided, including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the preceding embodiments.

[0008] In a fourth aspect of the invention, a non-transitory computer-readable storage medium is provided, the computer-readable storage medium storing instructions that, when executed, perform the method as described in any of the preceding claims.

[0009] In summary, one or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: This invention constructs a coupled observation data set by integrating infrared thermographic sequences and acoustic response sequences within a unified spatial registration framework. Temperature recovery delay and acoustic propagation disturbance characteristics are extracted separately to form a unified defect response feature set. This allows thermal and acoustic anomalies to collaboratively characterize structural changes caused by freezing and thawing under low-temperature cycling conditions within the same characterization space. Simultaneously, a coupled discriminant function enhances consistency and suppresses differences between the two types of physical responses. Furthermore, spatial continuity constraints are used to eliminate isolated noise interference and enhance the spatial distribution characteristics of real defects. Stage consistency correction suppresses fluctuations between different thermal response stages and observation cycles. Thus, a stable defect response distribution field is constructed under the joint constraints of multiple cycles, stages, and physical fields. Based on this distribution field, adaptive threshold mapping and connected region extraction are performed, enabling complex defects such as hidden cracks, insulation layer voids, and moisture migration to be continuously enhanced and stably identified even in low-temperature cycling environments, thereby significantly improving defect identification accuracy and discrimination stability. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the method for analyzing internal defects in building walls that integrates infrared thermal imaging and acoustic wave data, as disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of a normalization recovery curve disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of a module of the building wall internal defect analysis device that integrates infrared thermal imaging and acoustic wave data disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention.

[0011] Explanation of reference numerals in the attached drawings: 301, acquisition module; 302, processing module; 303, output module; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0013] In the description of the embodiments of the present invention, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0014] In the description of the embodiments of the present invention, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0015] While existing methods for detecting internal defects in building walls have incorporated infrared thermography and acoustic detection in low-temperature cycling environments, the lack of a unified multi-source data collaborative characterization mechanism and system constraints on spatial continuity and multi-stage response consistency makes it difficult to perform unified fusion analysis of temperature field anomalies and sound propagation anomalies under alternating freezing and thawing conditions. This hinders the simultaneous assurance of detection sensitivity and discrimination stability in identifying complex defects such as hidden cracks, insulation layer voids, and moisture migration, resulting in detection results that are easily affected by environmental disturbances and insufficient overall identification accuracy.

[0016] The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data disclosed in this invention is applied to a server. The server includes, but is not limited to, electronic devices such as mobile phones, tablets, wearable devices, and PCs (Personal Computers), and can also be a backend server running the method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0017] This embodiment discloses a method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data, referring to... Figure 1 It includes the following steps: S110: Obtain the infrared thermal image sequence and acoustic response sequence corresponding to the target wall and perform spatial registration processing to form a coupled observation data set.

[0018] S120, based on the coupled observation data set, construct the temperature recovery delay characterization quantity and the sound propagation disturbance characterization quantity to form a defect response feature set.

[0019] S130, constructs a coupled discriminant function for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response.

[0020] S140, based on the coupling discriminant function, spatial continuity constraint processing and stage consistency correction processing are performed to obtain a stable defect response distribution field.

[0021] S150 performs threshold mapping and region extraction processing based on a stable defect response distribution field, generates a set of defect regions, and outputs the internal defect identification results.

[0022] In one possible implementation, a temperature recovery delay characterization quantity and an acoustic propagation disturbance characterization quantity are constructed based on a coupled observation data set to form a defect response feature set. Specifically, this includes: performing observation unit partitioning on the coupled observation data set to obtain a regional observation mapping set; performing thermal image reference straightening on the infrared thermal image sequence based on the regional observation mapping set to obtain a regional thermal response trajectory set and a reference thermal response trajectory; constructing a temperature recovery delay characterization quantity based on the regional thermal response trajectory set and the reference thermal response trajectory; performing propagation path analysis and path disturbance straightening on the acoustic response sequence based on the regional observation mapping set to form a regional acoustic response trajectory set and a reference acoustic response trajectory; constructing an acoustic propagation disturbance characterization quantity based on the regional acoustic response trajectory set and the reference acoustic response trajectory; performing a screening process based on the temperature recovery delay characterization quantity and the acoustic propagation disturbance characterization quantity to form a regional defect candidate set; and binding the temperature recovery delay characterization quantity, acoustic propagation disturbance characterization quantity, and auxiliary constraint quantity corresponding to each region identifier in the regional defect candidate set to form a defect response feature set.

[0023] Specifically, a unified spatial coordinate system is first established around the target wall, and the pixel coordinates in the infrared thermal image sequence, the excitation position coordinates in the acoustic response sequence, and the sensor deployment coordinates are all mapped to this unified spatial coordinate system. Then, based on the preset grid scale, wall structure partition boundaries, and detection resolution requirements, the target wall is divided into observation units to form multiple observation sub-regions corresponding to different region identifiers. On this basis, each region identifier is simultaneously bound to its coverage area's thermal image pixel set, excitation response associated path set, and adjacent region relationships, thus forming a region observation mapping set. Here, the observation unit division process refers to discretizing the continuous wall detection area into the smallest spatial analysis unit that can be uniformly analyzed, so that subsequent thermal response analysis and acoustic response analysis are both carried out around the same region identifier. The region observation mapping set refers to the mapping structure that records each region identifier and its corresponding thermal image observation range, acoustic propagation range, neighborhood relationship, and time index relationship. To avoid overly coarse region division leading to the smoothing of minor defects, or overly fine region division leading to excessive noise, the region scale can be adaptively corrected based on the degree of local texture change and acoustic propagation path density. The expression for the region scale correction coefficient is:

[0024] in, This represents the scale correction factor for the region identified as r. The larger the value, the more fine the observation unit needs to be divided for that region. This indicates the intensity of thermal image texture changes within the area covered by the region identifier, and its value can be obtained from temperature gradient or grayscale co-occurrence statistics. This represents the maximum statistical value among the thermal image texture change intensities of each region of the current target wall. The effective sound propagation path density represents the density of sound propagation paths crossing the area marker, and its value is determined by the number of effective paths and the area of ​​the region. This represents the maximum statistical value among the effective sound propagation path densities in each area of ​​the current target wall. and This indicates the correction weight, the value of which is calibrated based on the requirements for thermal imaging detail and acoustic coverage; This represents a stable term, which is a small constant with a value greater than zero. This expression utilizes the local complexity of the thermal image and the coverage of the acoustic path to adjust the scale of the observation unit, so that the regional observation mapping set can maintain spatial consistency while adapting to the local non-uniform changes of the wall under freeze-thaw cycles.

[0025] After forming the regional observation mapping set, thermal image benchmark straightening is performed on the infrared thermal image sequence. This benchmark straightening process involves extracting the regional average temperature, regional peak temperature, regional temperature gradient, and time-varying temperature change curve from consecutive thermal image frames around each regional identifier. It also eliminates incomparable factors caused by initial temperature bias, differences in external radiation reflection, and environmental drift from different detection rounds, ensuring a unified comparison of thermal responses across different regions and time periods. Specifically, the original thermal response curves corresponding to each regional identifier are first extracted based on the regional observation mapping set. Then, reference regional identifiers are selected from regions with intact structures, stable responses, and stable performance across multiple rounds. Reference thermal response trajectories are constructed around these reference regional identifiers. Simultaneously, baseline correction and amplitude straightening are performed on the original thermal response curves corresponding to other regional identifiers, forming a set of regional thermal response trajectories. This set of regional thermal response trajectories refers to the set of temperature recovery trajectories for each regional identifier under a unified time axis and a unified relative temperature scale. The reference thermal response trajectory is the benchmark recovery trajectory formed after straightening the thermal response curves corresponding to the reference regional identifiers, used to characterize the normal thermal recovery behavior of an intact wall in the current detection scenario. The expression for the straightened recovery curve is:

[0026] in, The normalized thermal response curve for the region identified as r typically ranges between zero and one, reflecting the relative temperature evolution of the region during the recovery phase. This indicates the average temperature of the region at time t. This indicates the start time of the thermal response corresponding to the area identifier, and its value is determined by the first time point when the temperature change rate exceeds a preset threshold. This indicates the peak time of the thermal response corresponding to the region identifier, and its value is determined by the local maximum temperature point; , , and These represent the normalized thermal response curve, average temperature of the region, thermal response start time, and thermal response peak time corresponding to the reference region identifier, respectively. This normalization method transforms the absolute temperature differences between different regions into uniform recovery trajectory differences, enabling subsequent temperature recovery delay characterization quantities to focus on the thermal inertia changes caused by defects, rather than spurious differences caused by surface material color differences or ambient temperature drift.

[0027] After obtaining the set of regional thermal response trajectories and the reference thermal response trajectory, a temperature recovery delay characterization quantity is constructed based on both. Here, the temperature recovery delay characterization quantity refers to a comprehensive feature quantity used to quantify the time lag, trajectory deviation, and velocity hysteresis exhibited by a regional thermal recovery process relative to the reference thermal recovery process. Specifically, in the construction process, multiple recovery ratio nodes are first identified, such as time points when the recovery amplitude reaches 20%, 50%, 80%, and 90%, forming a regional recovery node sequence and a reference recovery node sequence. Then, the node delay component, trajectory deviation component, and slope hysteresis component are calculated. Finally, the target temperature recovery delay characterization quantity is obtained through weighted fusion. The node delay component reflects the arrival time deviation at different recovery stages; the trajectory deviation component reflects the cumulative difference in curve shape within the entire recovery interval; and the slope hysteresis component reflects the degree of anomaly in the recovery velocity variation pattern. The expression for the temperature recovery delay characterization quantity is:

[0028] in, The value represents the temperature recovery delay characteristic of the region identified as r. The larger the value, the more significant the thermal recovery anomaly in that region. K represents the number of recovery ratio nodes, and its value is determined by the preset number of recovery ratio nodes. This represents the node weight of the k-th recovery ratio node. Its value satisfies the condition that the sum of the weights of all nodes is one, and it is used to adjust the contribution ratio of different recovery stages in the result. The normalized node delay component of the region identifier r at the k-th recovery ratio node is represented by the value of the target region and the reference region at that recovery ratio node. The trajectory offset component, identified as region r, is obtained by averaging the absolute deviations of the normalized thermal response curve and the reference thermal response curve within the common recovery interval. The slope hysteresis component, identified as region r, is obtained by averaging the instantaneous recovery slope differences of the two curves within the common recovery interval. , and The fusion weight is determined by calibrating the contribution of node delay, overall trajectory offset, and recovery speed differences to defect identification. This expression can simultaneously capture the heat conduction path delay effects caused by insulation layer voids, hidden cracks, and localized water-bearing areas under freeze-thaw cycles, thus providing a stable thermal anomaly characterization basis for subsequent thermoacoustic fusion.

[0029] After forming the temperature recovery delay characterization quantity, propagation path analysis and path perturbation correction are performed on the acoustic response sequence based on the regional observation mapping set to form a set of regional acoustic response trajectories and a reference acoustic response trajectory. Here, propagation path analysis refers to determining the crossing relationship, coverage relationship, and main interaction relationship between each effective propagation path and each regional marker based on the excitation location, receiver location, regional coordinate relationship, and acoustic propagation coverage. Path perturbation correction refers to performing distance compensation, excitation intensity compensation, and frequency band consistency processing on the propagation time, attenuation degree, spectral centroid, and reflection peak characteristics of different paths, so that the acoustic response characteristics of different paths can be uniformly compared. Specifically, the arrival time of the first wave, the arrival time of the main reflected wave, the envelope energy curve, the spectral distribution, and the phase perturbation sequence are first extracted from the acoustic response sequence. Then, based on the geometric correspondence between the propagation path and the regional marker, these acoustic features are aggregated under each regional marker to form a set of regional acoustic response trajectories. Simultaneously, reference acoustic response features are extracted from the complete structural region corresponding to the reference regional marker to form a reference acoustic response trajectory. Here, the set of regional acoustic response trajectories refers to the set of acoustic response evolutions of each regional identifier under uniform propagation compensation and uniform characteristic scale; the reference acoustic response trajectory refers to the benchmark of normal acoustic propagation behavior corresponding to the reference regional identifier. The expression for the equivalent attenuation coefficient is:

[0030] in, This represents the equivalent attenuation coefficient of the region identifier r on the m-th effective propagation path. The larger the value, the more significant the sound energy attenuation. This represents the equivalent propagation length of the m-th effective propagation path within the region identified as r, and its value is calculated from the path geometric projection relationship. This represents the initial envelope amplitude of the m-th effective propagation path near the excitation starting point; This represents the effective envelope amplitude of the m-th effective propagation path after it traverses the region marked as r; This indicates the stability term. This expression incorporates the path length factor into the sound energy attenuation calculation, making the attenuation characteristics comparable under different propagation distances. This is beneficial for distinguishing between sound wave anomalies caused by real internal defects and normal attenuation caused by simple path length changes in low-temperature environments.

[0031] After obtaining the set of regional acoustic response trajectories and the reference acoustic response trajectory, an acoustic propagation disturbance characterization quantity is constructed based on both. Here, the acoustic propagation disturbance characterization quantity refers to a feature quantity used to quantify the comprehensive degree of anomaly exhibited by a certain region relative to the reference region in terms of propagation time delay, energy attenuation, and spectral shift. Specifically, for each region identifier, the propagation time difference component, attenuation disturbance component, and frequency band centroid shift component are extracted from all effective propagation paths. These components are then averaged within the path and fused between paths to form the acoustic propagation disturbance characterization quantity for the corresponding region identifier. The propagation time difference component mainly reflects the sound velocity anomaly caused by continuous material changes or the presence of voids; the attenuation disturbance component mainly reflects the energy loss anomaly caused by enhanced scattering, increased interface energy dissipation, or local loose structures; the frequency band centroid shift component mainly reflects the intensified attenuation of high-frequency components or spectral distortion. The expression for the acoustic propagation disturbance characterization quantity is:

[0032] in, This represents the acoustic propagation disturbance characteristic of the region identified as r. The larger the value, the more significant the acoustic propagation anomaly in that region. This represents the number of valid propagation paths corresponding to region identifier r, and its value is determined by the path filtering results. This represents the effective propagation time difference of region identifier r on the m-th effective propagation path, and its value is obtained from the time difference between the arrival time of the first wave and the excitation time. This indicates the effective propagation time difference of the reference area identifier along the corresponding path; and These represent the equivalent attenuation coefficients of region identifier r and reference region identifier on the m-th path, respectively. The region identifier is the centroid of the frequency band on the m-th path, and its value is obtained by power spectrum distribution statistics. This indicates the centroid of the frequency band of the reference area identifier along the corresponding path; , and The fusion weight is determined based on the contribution of propagation delay anomalies, attenuation anomalies, and spectral anomalies to defect identification. This expression unifies multiple acoustic anomalies caused by internal wall cracks, voids, and moisture migration in low-temperature cycling environments into a single comprehensive index, providing conditions for subsequent collaborative screening and fusion discrimination with the temperature recovery delay characterization.

[0033] After generating the temperature recovery delay characterization and acoustic propagation disturbance characterization, a screening process is performed based on these two parameters to form a candidate set of regional defects. This screening process involves retaining regions that may correspond to real internal defects based on the co-occurrence relationship, spatiotemporal stability relationship, and neighborhood support relationship between thermal and acoustic anomalies on the same regional identifier, while suppressing pseudo-anomaly regions that appear only in a single physics field or only momentarily in a single cycle. Specifically, the temperature recovery delay characterization and acoustic propagation disturbance characterization are first standardized and their deviations are calculated. Then, it is determined whether a region simultaneously satisfies both thermal and acoustic anomaly conditions. Furthermore, the consistency condition of consecutive observation cycles and the spatial neighborhood connectivity support condition are introduced to screen out the candidate set of regional defects. Here, the candidate set of regional defects refers to the set of regional identifiers that satisfy multi-physics field cooperative anomalies and possess a certain degree of spatiotemporal stability. The expression for the candidate determination coefficient is:

[0034] in, This represents the candidate decision coefficient for region identifier r. The larger the value, the more likely the region is to be a true defect region. This represents the temperature recovery delay characteristic of the region identified as r; and These represent the mean statistic and the dispersion statistic, respectively, of the overall temperature recovery delay characteristic. The quantity representing the acoustic propagation disturbance with region identifier r; and These represent the mean statistic and the dispersion statistic, respectively, of the overall sound propagation disturbance characteristics. This indicates the number of consecutive observation rounds in which the region identifier r simultaneously meets the conditions for both thermal and acoustic anomalies. This indicates the maximum number of common anomaly rounds among all region identifiers; This represents the number of neighborhood support for region identifier r, and its value is the number of regions in the neighborhood that simultaneously meet the candidate conditions. Indicates the maximum number of neighboring regions supported among all region identifiers; , , and This represents the screening weight, whose value is determined based on the contribution of thermal anomaly significance, acoustic anomaly significance, cycle stability, and spatial support to the identification of true defects. This expression can effectively eliminate false anomaly regions caused by surface condensation, localized reflection, single-excitation disturbance, or localized mechanical noise under freeze-thaw cycle conditions, making the candidate set of regional defects closer to the actual internal defect distribution.

[0035] After obtaining the candidate set of regional defects, the temperature recovery delay characterization, acoustic propagation disturbance characterization, and auxiliary constraint quantity corresponding to each regional identifier are bound together to form a defect response feature set. Here, binding means that around the same regional identifier, the same time identifier, and the same spatial location, the two types of core characterization quantities and multiple types of auxiliary constraint quantities are written into a unified feature structure, so that subsequent construction of the coupled discriminant function, spatial continuity constraint processing, and stage consistency correction processing can all be directly carried out around this unified feature structure. Here, auxiliary constraint quantities refer to constraint features other than the temperature recovery delay characterization and acoustic propagation disturbance characterization quantities, used to reflect the regional context state, enhance the ability to suppress false anomalies, and improve the reliability of fusion discrimination. These preferably include regional temperature gradient dispersion, the number of inflection points of the recovery curve, the number of path reflection peaks, the degree of skewness of the main frequency band, the regional neighborhood consistency coefficient, and the round stability coefficient. The defect response feature set is a set composed of feature records corresponding to multiple regional identifiers. Each feature record contains at least a regional identifier, a time identifier, a temperature recovery delay characterization quantity, an acoustic propagation disturbance characterization quantity, and auxiliary constraint quantities. The regional neighborhood consistency coefficient in the auxiliary constraint quantities can be expressed as:

[0036] in, This represents the neighborhood consistency coefficient of the region identified as r. The larger the value, the more consistent the region is with its neighbors in terms of thermal and acoustic anomalies. This represents the number of regions in the neighborhood set of region identifier r; This represents the set of spatial neighborhoods identified by region r; This represents the temperature recovery delay characteristic of the neighborhood region identified as j; This expression represents the acoustic propagation disturbance characteristic of the neighborhood region identified as j. It explicitly incorporates the cooperative consistency of thermoacoustic anomalies in the neighborhood into the auxiliary constraint, ensuring that the defect response feature set not only preserves the local anomaly intensity of a single region but also the support relationship of the surrounding region for that anomaly. This provides more complete contextual information for the subsequent construction of the coupled discriminant function. The defect response feature set formed through the above binding process can directly support the unified fusion representation of infrared thermography and acoustic response in low-temperature cyclic building wall scenarios, laying the foundation for the stable identification of complex internal defects.

[0037] In one possible implementation, a temperature recovery delay characterization quantity is constructed based on the regional thermal response trajectory set and the reference thermal response trajectory. Specifically, this includes: performing recovery stage identification on the regional thermal response trajectory set and the reference thermal response trajectory to form a regional recovery time period set; performing temperature trajectory normalization on the regional thermal response trajectory and the reference thermal response trajectory using the regional recovery time period set to obtain a normalized recovery curve; extracting the regional recovery node sequence and the reference recovery node sequence corresponding to the recovery ratio nodes based on the normalized recovery curve; constructing node delay components based on the regional recovery node sequence and the reference recovery node sequence; constructing trajectory offset components and slope hysteresis components based on the normalized recovery curve; and constructing a temperature recovery delay characterization quantity based on the node delay components, trajectory offset components, and slope hysteresis components.

[0038] Specifically, recovery phase identification is first performed on the regional thermal response trajectory corresponding to each region identifier and the reference thermal response trajectory to form a set of regional recovery time periods. In this embodiment, recovery phase identification refers to determining the effective time interval for the temperature response to fall back from the peak state to the stable background state after the thermal excitation ends or the external temperature disturbance weakens. This ensures that the construction of subsequent temperature recovery delay characterization quantities is always limited to the temperature recovery process with clear physical meaning, without mixing in the heating segment, plateau segment, or abnormal jump segment. Specifically, the regional thermal response trajectory is first smoothed on a unified time axis to reduce high-frequency disturbances caused by thermal imaging sampling noise and local reflections. Then, the first-order and second-order rates of change of temperature are calculated to identify the thermal response start time, thermal response peak time, and thermal response end time. The thermal response start time refers to the time point when the regional temperature begins to deviate from the baseline and change continuously; the thermal response peak time refers to the time point when the regional temperature reaches the highest response value before recovery; the thermal response end time refers to the time point when the regional temperature recovers to the stable fluctuation range and no longer shows significant deviation within a preset duration. To ensure comparability between the regional thermal response trajectory and the reference thermal response trajectory, both need to be identified separately at their respective recovery stages before being mapped onto a unified recovery stage representation framework. The recovery stage identification can be determined by the temperature change rate using the following formula:

[0039] in, This represents the rate of temperature change of region r at time t, and its value reflects how fast the temperature changes within a unit sampling interval. This represents the average temperature of the region identified as r at time t, and its value is derived from the region's thermal response trajectory. This represents the time interval between adjacent sampling moments, and its value is determined by the thermal imaging sampling frequency. The temperature change rate can be used to identify the direction and magnitude of temperature change as it enters the recovery phase, and combined with continuous multi-point consistency constraints, a set of regional recovery time periods is formed. The set of regional recovery time periods refers to the set recording the start time, peak time, end time, and corresponding time interval of each region identifier within the recovery phase, used for unified constraints on all subsequent recovery trajectories.

[0040] After forming a set of regional recovery time periods, temperature trajectory normalization is performed on the regional thermal response trajectories and the reference thermal response trajectory using this set to obtain normalized recovery curves. In this embodiment, temperature trajectory normalization refers to converting the original temperature curves formed by different regions due to differences in initial temperature, peak temperature, and recovery duration into relative recovery curves with a unified baseline, amplitude, and comparison scale. This ensures that recovery differences directly reflect differences in the thermal inertia within the wall, rather than differences in surface material color, ambient temperature bias, or detection starting point deviation. Specifically, the recovery time period identified for each region is first extracted based on the set of regional recovery time periods. Then, the temperature corresponding to the start of the thermal response is used as the baseline temperature, and the temperature difference between the peak thermal response time and the baseline temperature is used as the normalization amplitude. Baseline subtraction and amplitude normalization are then performed on the recovery curves. Simultaneously, the same processing is performed on the reference thermal response trajectory to form a reference normalized recovery curve.

[0041] Reference Figure 2 The reference normalized recovery curve reflects the standard thermal response behavior of the structurally intact region during the recovery phase. Its characteristics include timely recovery initiation, rapid recovery rate, and gradual stabilization. In contrast, the target region's normalized recovery curve shifts to the right relative to the reference curve, with a decreasing upward slope, indicating a significant lag in all recovery stages. The horizontal reference line in the figure corresponds to a preset recovery ratio node, used to determine the region's recovery node sequence and the reference recovery node sequence, thereby calculating the node delay component. The area difference between the two curves reflects the trajectory offset component, while the slope difference reflects the slope hysteresis component. These multidimensional differences are used to comprehensively construct a temperature recovery delay characterization, providing a unified and stable basis for subsequent fusion and discrimination with acoustic propagation disturbance characterization. The expression for the normalized recovery curve is:

[0042] in, The normalization recovery curve represents the region identified as r. Its value is usually between zero and one, reflecting the relative recovery status of the region during the recovery phase. This represents the original average temperature of the region identified as r at time t. This indicates the start time of the thermal response in the region identified as r, and its value is derived from the identification results during the recovery phase. This indicates the peak time of the thermal response in the region identified as r, and its value is derived from the identification results during the recovery phase. , , and These represent the normalized recovery curve, the original region average temperature, the thermal response start time, and the thermal response peak time corresponding to the reference thermal response trajectory, respectively. This represents a stable term, a small constant with a value greater than zero, used to avoid numerical anomalies caused by an excessively small denominator during the rounding process. Through temperature trajectory rounding, the subsequent comparison object changes from absolute temperature values ​​to differences in the shape and speed of the recovered trajectory, making the recovery behavior between different detection areas directly comparable.

[0043] After obtaining the normalized recovery curve, the regional recovery node sequence and reference recovery node sequence corresponding to the recovery ratio nodes are extracted based on the normalized recovery curve. In this embodiment, the recovery ratio node refers to several proportional positions with unified relative recovery significance selected on the normalized recovery curve, such as the node positions corresponding to 20%, 50%, 80%, and 90% of the peak amplitude, used to discretize the continuous recovery process into multiple key moments with clear stage meanings. The regional recovery node sequence refers to the set of arrival times corresponding to each regional identifier at the preset recovery ratio node; the reference recovery node sequence refers to the set of arrival times corresponding to the reference thermal response trajectory at the same recovery ratio node. In specific implementation, the recovery ratio set is first set, and then the time point for the first time reaching the corresponding ratio value is searched on each normalized recovery curve. If the sampling points do not fall precisely at the ratio value position, the precise arrival time can be determined by interpolation between adjacent sampling points. The expression for the regional recovery node time is:

[0044] in, This represents the recovery arrival time of region identifier r at the k-th recovery ratio node, and its value is determined by the position where the normalized recovery curve first reaches the corresponding recovery ratio. This represents the recovery arrival time of the reference thermal response trajectory at the k-th recovery ratio node; This represents the k-th recovery ratio node, whose value is between zero and one, and is determined by a preset set of recovery ratios; and These represent the end times of the thermal response for regions identified as r and the reference thermal response trajectory, respectively, and their values ​​are derived from the recovery phase identification results. By using the region recovery node sequence and the reference recovery node sequence, the continuous recovery trajectory can be converted into several stage-based time markers, providing a clear discrete basis for subsequent delay difference analysis.

[0045] After obtaining the regional recovery node sequence and the reference recovery node sequence, a node delay component is constructed based on both. In this embodiment, the node delay component refers to the degree of recovery arrival time lag of a certain region relative to the reference thermal response trajectory at each recovery ratio node, reflecting the delay differences at different stages of early, middle, and late recovery. Specifically, the arrival time difference between the target region and the reference region is calculated for each recovery ratio node, and then normalized using the relative scale of the reference recovery process to ensure uniform comparability of delay results under different detection rounds and different total recovery durations. The expression for the node delay component is:

[0046] in, This represents the node delay component of region r at the k-th recovery ratio node. The larger the value, the more significant the recovery lag in the corresponding recovery stage of the region. This indicates the recovery arrival time of region identifier r at the k-th recovery ratio node; This represents the recovery arrival time of the reference thermal response trajectory at the k-th recovery ratio node; Indicates the start time of the thermal response of the reference thermal response trajectory; The term represents the stable term. The nodal delay component can be used to quantitatively characterize the local hysteresis features at different recovery stages. For example, some degaussed regions may show obvious hysteresis in the early stage of recovery, while some water-bearing regions may show more significant delay in the later stage of recovery. This stage difference can be clearly distinguished by the nodal delay component.

[0047] Using only the node delay component is insufficient to fully describe the anomalous characteristics of the entire recovery trajectory. Therefore, it is necessary to further construct trajectory offset and slope hysteresis components based on the normalized recovery curve. In this embodiment, the trajectory offset component refers to the cumulative morphological deviation between the target region's normalized recovery curve and the reference normalized recovery curve within the entire common recovery interval, reflecting the continuous offset characteristics of the overall recovery process. In this embodiment, the slope hysteresis component refers to the cumulative difference in instantaneous recovery speed between the two normalized recovery curves within the common recovery interval, reflecting the anomalous degree of the recovery speed change pattern. Specifically, the recovery time interval jointly covered by the target region and the reference region is first determined. Then, the absolute difference between the two normalized recovery curves within the interval is integrated and averaged to form the trajectory offset component; simultaneously, the difference in the first derivative of the two curves is integrated and averaged to form the slope hysteresis component. The expression for the trajectory offset component is:

[0048] The expression for the slope hysteresis component is: in, The value of r represents the trajectory offset component of the region. The larger the value, the more significant the deviation between the target region and the reference region in the overall restored shape. The slope hysteresis component, denoted as r, represents the region. The larger the value, the more significant the difference in the recovery speed variation between the target region and the reference region. and This represents the start and end times of the common recovery interval between the target region and the reference region, and its value is determined by the time overlap interval of the two recovery curves. and These represent the target area normalized recovery curve and the reference normalized recovery curve, respectively, which are confined within the common recovery interval. and These represent the instantaneous recovery slopes of the two normalized recovery curves within the common recovery interval, and their values ​​can be obtained by differentiating adjacent sampling points. The trajectory offset component mainly characterizes the continuous deviation throughout the process and can reveal phenomena such as enhanced local thermal inertia or hindered thermal diffusion; the slope hysteresis component mainly characterizes the differences in the variation of recovery speed and can identify uneven recovery speed caused by differences in defect depth, water content, or internal interface state.

[0049] After obtaining the node delay component, trajectory offset component, and slope hysteresis component, a temperature recovery delay characterization quantity is constructed based on these three components. In this embodiment, the temperature recovery delay characterization quantity refers to a comprehensive thermal anomaly index obtained by unifying and fusing the local time delay in the recovery stage, the trajectory offset throughout the entire process, and the slope hysteresis in the recovery velocity around the same region identifier. This index is used subsequently to form a defect response feature set together with the acoustic propagation disturbance characterization quantity. Specifically, the node delay components at each recovery ratio node are first weighted and summarized according to stage importance, and then combined with the trajectory offset component and the slope hysteresis component according to the fusion weight to obtain the temperature recovery delay characterization quantity of the target region. The expression for the temperature recovery delay characterization quantity is:

[0050] in, The value represents the temperature recovery delay characteristic of the region identified as r. The larger the value, the more significant the thermal recovery anomaly in the region. K represents the number of recovery ratio nodes, and its value is determined by the number of nodes in the recovery ratio set. This represents the node weight corresponding to the k-th recovery ratio node. Its value satisfies that the sum of the weights of all nodes is one, and it is used to adjust the contribution ratio of different recovery stages in the overall result. This represents the node delay component with region identifier r at the kth recovery ratio node; This represents the trajectory offset component with region identifier r; This represents the slope hysteresis component of the region identified as r; , and The fusion weight is determined based on the contribution of stage delay, overall trajectory deviation, and abnormal recovery speed to the identification of internal wall defects. Through this comprehensive construction method, the temperature recovery delay characterization quantity retains both the hysteresis differences at different stages of the recovery process and the differences in morphology and speed evolution throughout the entire process. This allows for a more stable characterization of thermal recovery anomalies caused by complex internal defects such as hidden cracks, insulation layer voids, and moisture migration under low-temperature cycling conditions.

[0051] In one possible implementation, a coupling discriminant function is constructed for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response. Specifically, this includes: constructing thermal deviation components, acoustic deviation components, and constraint deviation vectors based on a region-normalized feature set, wherein the region-normalized feature set is formed by performing feature normalization processing on the defect response feature set; constructing a thermo-acoustic consistent coupling component and a thermo-acoustic difference suppression component based on the thermal and acoustic deviation components, and simultaneously constructing a region confidence coefficient based on the constraint deviation vector; constructing a coupling discriminant function based on the thermal deviation component, acoustic deviation component, thermo-acoustic consistent coupling component, thermo-acoustic difference suppression component, and region confidence coefficient to obtain a coupling discriminant value; performing temporal accumulation processing on the coupling discriminant value according to consecutive observation rounds to form a temporal coupling discriminant value; performing neighborhood collaborative enhancement processing based on the temporal coupling discriminant value to form a region fusion characterization value; and performing discriminant interval mapping processing based on the region fusion characterization value to form a fusion response probability.

[0052] Specifically, feature rounding is first performed on the defect response feature set to form a region-rounded feature set. In this embodiment, feature rounding refers to mapping the temperature recovery delay characterization, sound propagation disturbance characterization, and auxiliary constraint quantity from their original value spaces to a unified comparable value space around the same region identifier. This ensures that the subsequent construction of the coupling discriminant function is no longer directly affected by the numerical range, statistical dispersion, and differences in the distribution of abnormal extreme values ​​of different feature quantities. Specifically, based on the feature statistical results of all region identifiers within consecutive observation rounds, robust center extraction, robust discretization extraction, anomaly truncation, and interval compression are performed on each type of feature. The processing results are then written into the region-rounded feature set corresponding one-to-one with the region identifier. Here, the regional normalization feature set refers to a unified feature structure that simultaneously stores normalized temperature recovery delay, normalized acoustic propagation disturbance, and normalized auxiliary constraints under the same regional identifier. Auxiliary constraints are supplementary features used to describe the regional context state, trajectory stability, neighborhood consistency, and frequency band structure characteristics, such as regional temperature gradient dispersion, number of inflection points on the recovery curve, number of path reflection peaks, main frequency band skewness, round stability coefficient, and regional neighborhood consistency coefficient. To reduce the influence of extremely high-value regions on the overall normalization scale, a robust standardization method is preferred for normalizing various features. The normalization expression for a certain type of feature is:

[0053] in, This represents the qth type of normalized feature component of the region identified as r, and its value reflects the relative deviation of the region from the qth type of feature. This represents the original feature value of the qth class with region identifier r. Its value can be a temperature recovery delay characterization quantity, a sound propagation disturbance characterization quantity, or a certain auxiliary constraint quantity. This represents the median statistic, whose value is used to characterize the robust centrality level of this type of feature; The corresponding robust discrete estimate is used to characterize the robust fluctuation range of this type of feature; This represents a stable term, which is a small constant with a value greater than zero. Through this process, all features are compressed into a unified relative deviation space, facilitating the subsequent continuous construction of thermal deviation components, acoustic deviation components, and constraint deviation vectors.

[0054] After obtaining the region-aligned feature set, thermal deviation components, acoustic deviation components, and constraint deviation vectors are constructed around the region-aligned feature set. In this embodiment, the thermal deviation component refers to the degree of normalized anomaly of the region identifier relative to the reference background distribution in the dimension of temperature recovery delay characterization; the acoustic deviation component refers to the degree of normalized anomaly of the region identifier relative to the reference background distribution in the dimension of sound propagation disturbance characterization; and the constraint deviation vector refers to the multidimensional deviation set formed by each auxiliary constraint quantity relative to the corresponding reference background distribution, used for subsequent construction of region confidence coefficients. In specific implementation, reference background samples are first extracted from the low-response, low-fluctuation, and high-stability background regions to establish reference center parameters and reference discrete parameters for the temperature recovery delay characterization quantity, sound propagation disturbance characterization quantity, and each auxiliary constraint quantity. Then, the deviation values ​​of the thermal deviation component, acoustic deviation component, and each dimension of auxiliary constraint quantity are calculated for each region identifier. Here, the reference background sample refers to the set of background region samples selected based on the stable and complete regions under the freeze-thaw cycle scenario, used to characterize the thermal and acoustic baseline of the normal state of the wall under the current detection round. The expressions for the thermal deviation component and the acoustic deviation component are:

[0055] in, This represents the thermal deviation component of the region identified by r. The larger the value, the more significant the anomaly in the thermal dimension of the region. This represents the temperature recovery delay normalization feature value for region identifier r; This represents the central statistic of the temperature recovery delay normalized eigenvalues ​​in the reference background sample; This represents the discrete statistics of the temperature recovery delay normalized feature value in the reference background sample; The acoustic deviation component, denoted by r, represents the region. The larger the value, the more significant the acoustic anomaly in that region. This represents the normalized characteristic value of acoustic propagation disturbance in the region identified as r; and Let represent the central statistic and discrete statistic of the normalized eigenvalues ​​of acoustic propagation disturbance in the reference background sample, respectively. The constraint deviation component corresponding to the q-th auxiliary constraint can be expressed as:

[0056] in, This indicates the degree of deviation of region identifier r from the qth auxiliary constraint quantity; The region identifier is represented by the normalized eigenvalue of r on the qth auxiliary constraint. and Let represent the central statistic and discrete statistic of the q-th auxiliary constraint in the reference background sample, respectively. Then, let all... By combining them in a predetermined order, a constraint deviation vector for region identification is formed. Through the above construction method, thermal anomalies, acoustic anomalies, and context constraint anomalies are all uniformly expressed as the degree of deviation relative to the reference background, thus providing a consistent input basis for subsequent thermo-acoustic coupling enhancement and difference suppression.

[0057] After obtaining the thermal deviation component, acoustic deviation component, and constraint deviation vector, a thermo-acoustic co-coupling component and a thermo-acoustic difference suppression component are constructed based on the thermal and acoustic deviation components. Simultaneously, a region confidence coefficient is constructed based on the constraint deviation vector. In this embodiment, the thermo-acoustic co-coupling component refers to the multiplicative coupling quantity that enhances the characterization of a dual-physics-field co-occurrence anomaly when a region exhibits significant anomalies in both the thermal and acoustic dimensions. In this embodiment, the thermo-acoustic difference suppression component refers to the difference quantity that weakens the characterization of an inconsistent anomaly when a region exhibits significant anomalies only in a single physics field while the response in the other physics field is insufficient. In this embodiment, the region confidence coefficient refers to the degree of confidence obtained by comprehensively evaluating whether the current region's characteristics conform to the spatial continuity, frequency band stability, trajectory smoothness, and cycle consistency of a true internal defect using the constraint deviation vector. Specifically, firstly, positive anomaly screening is performed on the thermal deviation component and the acoustic deviation component, retaining only the positive anomaly coupling portion where both are greater than zero, to avoid erroneous amplification of background fluctuations or reverse anomalies; then, the absolute difference between the two is calculated to measure the degree of inconsistency between thermal and acoustic anomalies; subsequently, the constraint deviation vector is mapped to a regional confidence coefficient through nonlinear compression and weighted aggregation, enabling multiple contextual constraints to jointly adjust the confidence level of the current region. The expressions for the thermoacoustic consistent coupling component and the thermoacoustic difference suppression component are:

[0058] in, The thermoacoustic coherent coupling component, denoted as r, indicates that the region has stable positive anomalies in both the thermal and acoustic dimensions. This represents the thermoacoustic difference suppression component, denoted as region r. A larger value indicates a higher degree of anomalous inconsistency in both the thermal and acoustic dimensions of the region. The expression for the region confidence coefficient is:

[0059] in, This represents the confidence coefficient of the region identified as r. Its value is between zero and one. The larger the value, the more the region matches the comprehensive performance of the real defect in the context constraints. represents the bias coefficient, whose value is used to adjust the overall confidence level baseline; Q represents the number of auxiliary constraints. This represents the modulation weight corresponding to the q-th auxiliary constraint, and its value is determined based on the contribution of this constraint to the ability to suppress pseudo-anomalies. Through this process, the common anomaly of the two physical fields can be explicitly enhanced, the pseudo-anomaly caused by their inconsistency can be explicitly weakened, and the credibility of the result can be further constrained using the regional confidence coefficient, so that the subsequent coupling discriminant function is more consistent with the actual response law of internal defects in low-temperature cyclic walls.

[0060] After obtaining the thermal deviation component, acoustic deviation component, thermo-acoustic consistent coupling component, thermo-acoustic difference suppression component, and regional confidence coefficient, a coupling discriminant function is constructed based on these quantities to obtain the coupling discriminant value. In this embodiment, the coupling discriminant function refers to a fusion function that integrates the contribution of a single thermal anomaly, the contribution of a single acoustic anomaly, the enhancement of a common thermo-acoustic anomaly, the suppression of thermo-acoustic inconsistency, and the modulation of contextual confidence into the same discriminant model, all centered around the same regional identifier. In this embodiment, the coupling discriminant value refers to the regional-level fusion discriminant result calculated through the coupling discriminant function, used to characterize the relative probability that the current region belongs to an internal defect region in the current observation round. Specifically, a linear principal term is first used to perform a basic weighting on the thermal deviation component and the acoustic deviation component. Then, the thermo-acoustic consistent coupling component is superimposed on the principal term as a cooperative enhancement term, while the thermo-acoustic difference suppression component is deducted from the principal term as a penalty term. Finally, the regional confidence coefficient is used to modulate the overall result. The expression of the coupling discriminant function is:

[0061] in, The coupling discriminant value represents the region identifier r. The larger the value, the more likely the region is to correspond to a real internal defect. This represents the weight of the thermal deviation component, and its value is used to adjust the contribution of the temperature recovery delay anomaly to the overall discrimination. This represents the weight of the acoustic deviation component, and its value is used to adjust the contribution of acoustic propagation disturbance anomalies to the overall discrimination. This represents the weight of the thermoacoustic coherent coupling component, and its value is used to enhance the discrimination gain brought about by the common anomaly of the two physics fields. This represents the weight of the thermoacoustic difference suppression component, and its value is used to weaken the discriminant value of the region where thermal anomalies and acoustic anomalies are inconsistent; This represents the confidence coefficient of the region identified as r. The core of this coupled discriminant function is that it not only preserves the independent discriminant information of infrared thermal images and acoustic responses, but also highlights the co-occurrence relationship of the two types of information in the same region. At the same time, it uses context constraints to reduce the influence of false anomalies caused by surface frost, local reflections, single acoustic disturbances, or edge occlusion, thereby forming a more stable fusion characterization basis.

[0062] After obtaining the coupling discrimination value, a time-series accumulation process is performed on the coupling discrimination value according to the continuous observation rounds to form a time-series coupling discrimination value. In this embodiment, the continuous observation rounds refer to multiple coupling observations and fusion discrimination processes formed around the same target wall at adjacent detection times, adjacent freeze-thaw stages, or adjacent excitation rounds. In this embodiment, the time-series accumulation process refers to recursively fusing the coupling discrimination results in historical observation rounds with the coupling discrimination results in the current observation round, so that the stable defect areas gradually increase in time, while the noise areas that only occasionally appear in a single round gradually decrease. In this embodiment, the time-series coupling discrimination value refers to the regional discrimination result after fusing historical time-series information. In specific implementation, the time-series coupling discrimination value of the current region in the previous observation round is used as the historical item, and the coupling discrimination value of the current round is used as the new input item, and an exponential recursive method is used for updating. The expression for the time-series accumulation process is:

[0063] in, The region identifier is the temporal coupling discriminant value of r in the nth observation round; This indicates that the region identifier is the temporal coupling discriminant value of r in the previous observation round; The region identifier is the coupling discriminant value of r in the current observation round; This represents the time-preservation coefficient, with a value between zero and one, used to adjust the weight distribution between historical cumulative results and the current judgment result. Through this recursive update method, the thermoacoustic anomalies that the real internal defects continuously exhibit in multiple rounds can be stably preserved, while occasional anomalies caused by transient disturbances in the low-temperature environment, temporary surface attachments, or instability of a single excitation are significantly suppressed, thereby improving the stability of subsequent spatial processing.

[0064] After generating the temporal coupling discriminant value, neighborhood collaborative enhancement processing is performed based on the temporal coupling discriminant value to generate a regional fusion characterization value. In this embodiment, neighborhood collaborative enhancement processing refers to using the temporal coupling discriminant values ​​of other regional identifiers within the same spatial neighborhood to enhance support or suppress the boundary of the current region, thereby strengthening the spatial continuity of the defective region and weakening isolated, occasional high-value points. In this embodiment, the regional fusion characterization value refers to the spatial fusion result that integrates the temporal anomaly intensity, neighborhood collaborative support intensity, and boundary inconsistency degree of the region. Specifically, a spatial neighborhood set for each regional identifier is first established based on the regional observation mapping set. Then, a weighted average of the temporal coupling discriminant values ​​within the neighborhood is used to characterize the neighborhood support intensity. Simultaneously, the degree of boundary abrupt change between the current region and its neighboring regions is calculated to construct a boundary suppression term. Finally, the temporal coupling discriminant value, neighborhood support term, and boundary suppression term of the current region are combined according to predetermined weights to obtain the regional fusion characterization value. The expression for the regional fusion characterization value is:

[0065] in, The value represents the fusion representation of the region identified as r. The larger the value, the more likely the region is to be a true defect region under the combined effects of its own anomalies, neighborhood support, and boundary coordination. This represents the weight of the central region, and its value is used to adjust the contribution of the anomalies of the current region to the fusion result; This represents the neighborhood collaboration weight, whose value is used to adjust the contribution of neighborhood support to the fusion result; This represents the number of regions in the neighborhood set of region identifier r; This indicates the temporal coupling discriminant value of the neighborhood region identifier j in the current observation round; This represents the boundary incoordination component identified as r, whose value reflects the degree of abrupt change in the temporal coupling discriminant value between the current region and the surrounding regions. This represents the boundary suppression weight, whose value is used to suppress isolated high-value areas that are severely inconsistent with the surrounding area. This treatment is suitable for real defects in low-temperature cyclic walls, which often exhibit spatial characteristics of continuous distribution in sheet-like, band-like, or clump-like forms. It can enhance the stable response of the main defect area while maintaining the defect boundary.

[0066] After obtaining the regional fusion representation value, a discriminative interval mapping process is performed based on the regional fusion representation value to form the fusion response probability. In this embodiment, the discriminative interval mapping process refers to transforming the regional fusion representation value from an unbounded discriminative space to a unified probability space or a unified level space, enabling subsequent spatial continuity constraint processing, stage consistency correction processing, and defect region extraction processing to be carried out based on explicit probabilistic semantics. In this embodiment, the fusion response probability refers to the defect response probability corresponding to the regional identifier after considering comprehensive thermal anomalies, acoustic anomalies, temporal cumulative results, and neighborhood collaboration results. Specifically, a monotonic nonlinear compression method is preferably used to map the regional fusion representation value to a range between zero and one, so that the larger the regional fusion representation value, the closer the corresponding fusion response probability is to one. The expression for the fusion response probability is:

[0067] in, This represents the probability of the fusion response for the region identified as r. Its value is between zero and one. The larger the value, the more likely the region is to correspond to an internal defect. This represents the fusion representation value of the region identified as r. Using this mapping method, fusion results from different walls, different inspection rounds, and different environmental backgrounds can be compressed into a unified probabilistic scale, facilitating subsequent threshold mapping, connected component extraction, geometric consistency screening, and defect severity grading. Furthermore, the fusion response probability can be directly compared with historical round results to analyze the growth, decay, and stabilization trends of defect responses in the same region.

[0068] In one possible implementation, spatial continuity constraint processing and stage consistency correction processing are performed based on the coupled discriminant function to obtain a stable defect response distribution field. Specifically, this includes: generating an initial discriminant distribution field by identifying each region based on the coupled discriminant function; constructing a set of regional neighborhood weights based on the initial discriminant distribution field; performing spatial continuity constraint processing on the initial discriminant distribution field using the set of regional neighborhood weights to obtain spatial constraint discriminant values; performing boundary preservation enhancement based on the spatial constraint discriminant values ​​to obtain boundary preservation discriminant values; performing stage division based on the boundary preservation discriminant values ​​to form a stage division set and constructing a stage discriminant sequence; constructing stage consistency coefficients based on the stage discriminant sequence; performing stage consistency correction processing on the boundary preservation discriminant values ​​based on the stage consistency coefficients to obtain stage correction discriminant values; and performing multi-round cumulative fusion processing based on the stage correction discriminant values ​​to obtain stable defect response values ​​and map them to a stable defect response distribution field.

[0069] Specifically, an initial discrimination distribution field is first generated for each region identifier based on the coupled discriminant function. In this embodiment, the initial discrimination distribution field refers to the two-dimensional or three-dimensional discrimination field formed by mapping the coupled discriminant value corresponding to each region identifier to a unified spatial grid according to its spatial position relationship under the same observation round, which is used to express the initial defect response intensity distribution of each region on the wall surface or inside the wall. In specific implementation, the region-level coupled discriminant value output by the coupled discriminant function is first bound to the spatial coordinate information in the region observation mapping set, and then all region identifiers are sequentially filled into the spatial matrix according to the unified grid topology, while retaining the region adjacency relationship, boundary relationship and region scale information. In order to ensure spatial consistency between different observation rounds, the grid needs to be uniformly numbered and spatially aligned so that the same region identifier always corresponds to the same spatial position in different rounds. The essence of this process is to reconstruct the discrete region-level discrimination result into a continuous spatial expression, providing an input basis for subsequent spatial continuity constraint processing.

[0070] After obtaining the initial discrimination distribution field, a regional neighborhood weight set is constructed based on the initial discrimination distribution field. In this embodiment, the regional neighborhood weight set refers to defining the spatial correlation strength between each regional identifier and its neighboring regional identifiers, used to characterize the degree of mutual influence between spatially adjacent regions. In specific implementation, a neighborhood topology relationship is first established based on the regional observation mapping set. Regions that share a boundary with the target region, are less than a preset threshold in distance, or have a physical connection are included in the neighborhood set. Then, the neighborhood weights are constructed by combining spatial distance, shared boundary length, and material consistency. The expression for the neighborhood weight is:

[0071] in, This represents the neighborhood weight between region r and region j. The larger the value, the stronger the spatial association between the two regions. This represents the spatial distance between the center points of two regions, and its value is calculated from the region coordinates. This represents the distance adjustment parameter, whose value is used to control the distance decay rate. This represents the length of the shared boundary between two regions, and its value reflects the degree of spatial contact. This represents the maximum shared boundary length among all neighborhoods of region identifier r; This represents the stable term. By simultaneously considering both distance and contact, the neighborhood weights reflect both geometric proximity and structural connectivity.

[0072] After obtaining the set of neighborhood weights, spatial continuity constraint processing is performed on the initial discrimination distribution field using the set of neighborhood weights to obtain spatially constrained discrimination values. In this embodiment, spatial continuity constraint processing refers to using neighborhood weights to weightedly fuse the coupled discrimination values ​​of each region identifier, thereby enhancing spatially continuous defect regions and suppressing isolated anomalies. Specifically, for each region identifier, the coupled discrimination values ​​of all regions within its neighborhood are weighted according to the neighborhood weights and fused with its own discrimination value to obtain the spatially constrained discrimination result. The expression for the spatially constrained discrimination value is:

[0073] in, This represents the spatial constraint discrimination value for region r, and its value reflects the comprehensive anomaly degree of the region under the synergistic effect of its neighborhood. This represents the coupling discriminant value for the neighboring region identified as j; This represents the set of neighborhoods of region denoted as r. The principle behind this process is that real defects are usually spatially continuous, and neighborhood weighting can enhance continuous regions, while random noise is weakened due to the lack of neighborhood support.

[0074] After obtaining the spatial constraint discrimination value, boundary preservation enhancement is performed based on the spatial constraint discrimination value to obtain the boundary preservation discrimination value. In this embodiment, boundary preservation enhancement refers to, after spatial smoothing, to avoid excessive blurring of the defect boundary, difference preservation is performed on the boundary region with large discriminant value changes, so that the defect region boundary remains clear. In specific implementation, the boundary position is identified by calculating the region discrimination gradient, and a boundary preservation factor is introduced to modulate the smoothing result. The expression of the boundary preservation factor is:

[0075] The expression for the boundary preservation criterion value is: in, This represents the boundary preservation factor for the region identified as r. The smaller the value, the more likely the region is located at the boundary. The gradient represents the spatial constraint discrimination value, and its value is obtained by neighborhood difference calculation; Indicates the gradient adjustment parameter; This indicates that the boundary maintains the discriminant value; This represents the original coupling discriminant value. This processing avoids over-smoothing of the boundary by adding weights to the original discriminant value in the boundary region.

[0076] After obtaining the boundary preservation discrimination values, stage division is performed based on these values ​​to form a stage set and construct a stage discrimination sequence. In this embodiment, stage division refers to dividing the entire observation process into multiple stages with similar physical characteristics, such as a heating stage, a stabilization stage, and a cooling stage, based on the thermal response process or the detection time series. The stage discrimination sequence refers to the sequence of discrimination values ​​corresponding to the same region at different stages. In specific implementation, the stage boundaries are first determined based on the time axis or the trend of the thermal response curve, and then the corresponding boundary preservation discrimination values ​​are extracted within each stage to form a stage discrimination sequence, thereby achieving segmented analysis in the time dimension.

[0077] After obtaining the stage discrimination sequence, a stage consistency coefficient is constructed based on the stage discrimination sequence. In this embodiment, the stage consistency coefficient is an index that measures the stability of the discrimination results of a certain region in different stages, and is used to distinguish between persistently abnormal regions and staged noise regions. In specific implementation, the consistency index is constructed by statistically analyzing the dispersion and average level of the discrimination values ​​in different stages. The expression for the stage consistency coefficient is:

[0078] in, The phase consistency coefficient represents the region identified as r. The closer the value is to one, the more stable the region remains in different phases. This represents the boundary preservation discriminant value at stage k; K represents the number of stages. A statistic representing the degree of dispersion; This represents the mean statistic. This process identifies stability defects by comparing the degree of fluctuation between periods.

[0079] After obtaining the stage consistency coefficient, stage consistency correction is performed on the boundary preservation discriminant value based on the stage consistency coefficient to obtain the stage-corrected discriminant value. In specific implementation, the stage consistency coefficient is used as a weight to apply to the boundary preservation discriminant value, thereby enhancing cross-stage stable regions and suppressing stage-abnormal regions. The expression for the stage-corrected discriminant value is:

[0080] in, This represents the stage correction discriminant value for the region identified as r. This processing modulates the spatial discriminant result with temporal consistency, making the result more stable and reliable.

[0081] After obtaining the stage-corrected discriminant value, a multi-round cumulative fusion process is performed based on the stage-corrected discriminant value to obtain a stable defect response value and map it to a stable defect response distribution field. In this embodiment, the multi-round cumulative fusion process refers to recursively fusing the discriminant results across multiple observation rounds, gradually strengthening long-term stable defect regions while gradually weakening random fluctuations. The expression for the stable defect response value is:

[0082] in, This indicates the stable defect response value of region r under the nth round of observation; This indicates the result of the previous round; This indicates the current round stage correction discriminant value; This represents the historical retention coefficient. Finally, the stable defect response values ​​corresponding to all region identifiers are mapped according to their spatial locations, thus forming a stable defect response distribution field, which provides stable input for subsequent threshold mapping and defect region extraction.

[0083] In one possible implementation, before performing threshold mapping and region extraction processing based on the stable defect response distribution field to generate a defect region set and output the internal defect identification result, the method further includes: performing background sample screening based on the normalized response field to obtain a background candidate region set, wherein the normalized response field is obtained by performing response value normalization processing on the stable defect response distribution field; performing global background statistical modeling and global tail risk estimation based on the background candidate region set to form a global background center parameter, a global background discrete parameter, and a global tail boundary parameter; performing global threshold fusion processing based on the global background center parameter, the global background discrete parameter, and the global tail boundary parameter to form an initial global baseline threshold; and performing threshold fusion processing based on the normalized response field to generate a defect region set. The system performs local neighborhood modeling and local background denoising around each region's identifier to form a local background candidate sample set. Local robust statistical modeling is then performed on this set to generate local background center parameters, local background discrete parameters, and local tail boundary parameters. Local threshold fusion is then performed based on the local background center parameters, local background discrete parameters, and local tail boundary parameters, combined with the local complexity coefficient, to form an initial local baseline threshold. Correction processing is then performed based on the normalized response field to generate stage drift correction and structural partition correction. Finally, combining the stage drift correction and structural partition correction, a final correction is applied to the initial global baseline threshold and the initial local baseline threshold to form the global baseline threshold and the local baseline threshold.

[0084] Specifically, the stable defect response distribution field is first subjected to response value normalization processing to form a normalized response field. In this embodiment, response value normalization processing refers to compressing the stable defect response values ​​corresponding to each region identifier from the original discrimination value space to a unified relative response space, so that stable defect response values ​​under different wall types, different detection rounds, and different thermo-acoustic coupling intensities can be directly compared. Specifically, the stable defect response values ​​corresponding to all region identifiers in the current stable defect response distribution field are first extracted, and then the stable defect response values ​​of each region identifier are linearly normalized and mapped using the minimum and maximum stable defect response values ​​of the entire field as upper and lower bounds, thereby forming a normalized response field. Here, the normalized response field refers to the response field composed of the normalized response values ​​corresponding to each region identifier under a unified spatial coordinate system, which essentially reflects the relative anomaly strength of each region relative to the background of the entire wall. The expression for the normalized response value is:

[0085] in, This represents the normalized response value for the region identified as r. Its value is between zero and one, and the larger the value, the more significant the relative anomaly of the region. This represents the stable defect response value for the region identified as r, and its value is derived from the stable defect response distribution field. This represents the minimum stable defect response value in the current stable defect response distribution field; This represents the maximum stable defect response value in the current stable defect response distribution field; This represents a stable term, a small constant with a value greater than zero. This process provides a unified numerical scale for subsequent background sample screening and threshold statistical modeling.

[0086] After obtaining the normalized response field, background sample screening is performed based on the normalized response field to obtain a set of background candidate regions. In this embodiment, background sample screening refers to identifying regions from the normalized response field that are more likely to correspond to a complete wall background rather than defect anomalies, around each region identifier, to establish statistical benchmarks for subsequent global and local thresholds. Specifically, the normalized response value, regional neighborhood gradient value, regional neighborhood variance value, and regional stability indicator are first extracted around each region identifier. Then, based on the joint conditions of low response, low gradient, low fluctuation, and high stability, the background candidate evaluation is performed on each region identifier. Here, the set of background candidate regions refers to the set of region identifiers that are considered to be more consistent with the characteristics of a complete wall background after screening; the regional neighborhood gradient value refers to the intensity of the difference in normalized response between the current region and its neighboring regions, used to characterize the degree of spatial abrupt change; the regional neighborhood variance value refers to the dispersion of the normalized response value within the current region's neighborhood, used to characterize the degree of local background fluctuation; and the regional stability indicator refers to the degree to which the region maintains a stable response in consecutive observation rounds. The expression for the background candidate determination coefficient is:

[0087] in, This represents the background candidate determination coefficient for region identifier r. The larger the value, the more likely the region is to belong to the background region. This represents the normalized response value for the region identifier r; This represents the gradient value of the neighborhood of the region identified as r, and its value can be obtained from the difference in the normalized response between the current region and the neighboring regions. This represents the maximum statistical value of the neighborhood gradient values ​​in the entire field. This represents the variance of the neighborhood of the region identified as r; This represents the maximum statistical value of the neighborhood variance across the entire field. This represents the stability indicator for the region identified as r. A larger value indicates a more stable response across multiple rounds. , , and The weights are determined based on the contributions of low-response features, low-gradient features, low-fluctuation features, and high-stability features to background recognition. This screening process can eliminate potential defect areas, edge-perturbed areas, and local noise areas as much as possible, making the set of background candidate areas closer to the real and complete background.

[0088] After forming a set of background candidate regions, global background statistical modeling and global tail risk estimation are performed based on the set to form global background center parameters, global background discrete parameters, and global tail boundary parameters. In this embodiment, global background statistical modeling refers to robustly modeling the overall background center level and overall background fluctuation range of the normalized response field within the entire wall area, using the set of background candidate regions as statistical samples. Global tail risk estimation in this embodiment refers to estimating the upper boundary that natural background fluctuations may reach around the high-response tail of the background distribution, used to distinguish between high-end background fluctuations and true anomalous responses. Specifically, median statistics and median absolute deviation statistics are preferably used to construct the global background center parameters and global background discrete parameters, and high quantile statistics are used to construct the global tail boundary parameters. Here, the global background center parameter refers to the robust center level of the normalized response values ​​in the set of background candidate regions; the global background discrete parameter refers to the robust fluctuation range of the normalized response values ​​around the center level in the set of background candidate regions; and the global tail boundary parameter refers to the upper boundary of the normalized response values ​​at the high quantile in the set of background candidate regions. Their expressions are as follows:

[0089] in, Indicates the global background center parameter; Represents the discrete parameters of the global background; Indicates global tail boundary parameters; denoted as the set of candidate background regions; p represents the high quantile level, with a value between zero and one, preferably close to one, to characterize the upper bound of the high-side tail of the background. By employing robust statistical modeling, the influence of a small number of weak defect residual samples and boundary perturbation samples on the background statistical results can be reduced.

[0090] After obtaining the global background center parameters, global background discrete parameters, and global tail boundary parameters, a global threshold fusion process is performed based on these three parameters to form an initial global baseline threshold. In this embodiment, the global threshold fusion process refers to integrating the overall background center level, the overall background fluctuation range, and the high-side tail boundary of the background into an initial anomaly boundary threshold applicable to the entire wall, so that it reflects both the overall position of the background and the natural undulations of the background's upper edges. Here, the initial global baseline threshold refers to the global anomaly discrimination threshold generated solely based on the global background statistical characteristics before considering stage drift and structural partition offset. The expression for the initial global baseline threshold is:

[0091] in, This represents the initial global baseline threshold, whose value is used to characterize the global background anomaly boundary of the entire wall surface; Indicates the global background center parameter; Represents the discrete parameters of the global background; This represents the global discrete adjustment coefficient, whose value is used to control the degree of influence of background fluctuations on the global threshold. Indicates global tail boundary parameters; and The fusion weight is defined as the sum of the two values ​​being one. This fusion process avoids both relying solely on the mean and variance, which could lead to an excessively low threshold, and relying solely on high quantile boundaries, which could lead to an excessively high threshold, thus achieving a balance between sensitivity and robustness.

[0092] After establishing the initial global baseline threshold, local neighborhood modeling and local background denoising are performed around each region identifier based on the normalized response field to form a local background candidate sample set. In this embodiment, local neighborhood modeling refers to establishing a local statistical pane corresponding to the spatial location of each region identifier to capture the background undulation level, material non-uniformity, and local anomalous distribution state within the region's neighborhood. Local background denoising in this embodiment refers to removing high-response peak regions, strong gradient boundary regions, and suspicious regions with obvious anomalous connectivity support within the local statistical pane, retaining only samples that better match the characteristics of a complete local background. Here, the local background candidate sample set refers to the set of local background samples retained after background denoising within the local neighborhood of a certain region identifier. Specifically, the pane radius is first adaptively adjusted based on the basic pane radius, combined with the region neighborhood variance and material non-uniformity indicator, and then samples are filtered within the local pane according to low response, low gradient, and low connectivity support conditions. The expression for the local statistical pane radius is:

[0093] The expression for the local background filtering coefficient is: in, This represents the radius of the local statistical pane labeled r. A larger value indicates that a larger neighborhood is needed to find local background samples. Indicates the radius of the base pane; This represents the variance of the neighborhood of the region identified as r; The material non-uniformity indicator for the region denoted as r can be determined by a combination of thermal image texture differences, sound propagation path dispersion, or wall finish zoning results. and Indicates the pane adjustment weight; This represents the local background filtering coefficient for region j within the local statistics pane labeled r. This represents the normalized response value for region identifier j; This represents the gradient value of the neighborhood of region j; The connectivity support coefficient for region j is represented by its value, which reflects whether the region is in an abnormal connectivity structure. This represents the maximum statistical value of the overall connectivity support coefficient; , and This indicates the local background filtering weight. Through this process, each region identifier can have a set of local background candidate samples that match its own background environment.

[0094] After forming a local background candidate sample set, local robust statistical modeling is performed on the local background candidate sample set to form local background center parameters, local background discrete parameters, and local tail boundary parameters. In this embodiment, local robust statistical modeling refers to establishing the local background center level, fluctuation range, and upper high-side bound of the local background within its local background candidate sample set for each region identifier using robust statistical methods. Here, the local background center parameter refers to the robust center level of the local background samples surrounding a certain region; the local background discrete parameter refers to the robust fluctuation range of the local background samples surrounding a certain region; and the local tail boundary parameter refers to the upper boundary of the distribution of local background samples around a certain region at the high quantile. Their expressions are as follows:

[0095] in, This represents the local background center parameter for the region identified as r; This represents the discrete parameters of the local background, identified as region r. This represents the local tail boundary parameter of the region identified as r; This represents the set of local background candidate samples corresponding to region identifier r; This indicates a high quantile level. This region-by-region robust statistical modeling can accommodate the non-uniformity of different parts of the wall due to variations in material composition, insulation layer structure, heat reflection conditions, and sound propagation background.

[0096] After obtaining the local background center parameter, local background discrete parameter, and local tail boundary parameter, local threshold fusion is performed based on the combination of these three parameters and the local complexity coefficient to form an initial local baseline threshold. In this embodiment, the local complexity coefficient is a comprehensive index used to characterize the background complexity within a certain region's neighborhood, primarily reflecting the degree of local response fluctuation and material inhomogeneity. In this embodiment, local threshold fusion refers to enhancing the contribution of the center discrete threshold term in stable local background regions and enhancing the contribution of the tail distributed threshold term in complex local background regions. Here, the initial local baseline threshold refers to the local anomaly discrimination threshold generated based on local background characteristics before stage drift correction and structural partitioning correction are performed. The expressions for the local complexity coefficient and the initial local baseline threshold are:

[0097] in, This represents the local complexity coefficient of the region identified as r. The larger the value, the more complex the neighborhood background of that region. This represents the maximum statistical value of the neighborhood variance across the entire field. This represents the maximum statistical value indicating non-uniformity of materials across the entire field. This represents the local threshold adaptive weight for region identifier r, whose value is obtained by monotonically mapping the local complexity coefficient and is between zero and one. This represents the initial local baseline threshold for the region identified as r; This represents the local discrete adjustment coefficient. This adaptive fusion method allows the local threshold to maintain higher sensitivity in stable backgrounds and higher robustness in complex backgrounds.

[0098] After establishing the initial global baseline threshold and the initial local baseline threshold, a correction process is performed based on the normalized response field to generate the stage drift correction and the structural partition correction. In this embodiment, the correction process refers to scenario-based compensation of the thresholds to address the systematic background shift caused by changes in the detection stage and differences in wall structural partitions under freeze-thaw cycles. Here, the stage drift correction refers to the overall background response drift caused by different thermal recovery stages, different excitation cycles, or different observation periods; the structural partition correction refers to the local background shift caused by systematic differences in the heat transfer and sound transmission characteristics of different structural partitions of the wall. Specifically, the current detection data is first divided into stages based on the time axis and thermal response state. Then, the difference in the mean and dispersion of the normalized response field between the current stage and the reference stage is statistically analyzed to obtain the stage drift correction. Subsequently, the wall is structurally partitioned based on wall material partitions, insulation layer partitions, finishing partitions, or structural node partitions. The systematic shift between the normalized response field of each partition and the normalized response field of the entire wall is compared to obtain the structural partition correction. Its expression is:

[0099] in, Indicates the stage drift correction amount; This represents the average statistical value of the adjusted response values ​​corresponding to all region identifiers within the current stage; This represents the mean statistic of the adjusted response values ​​corresponding to all region identifiers within the reference phase. This represents a statistic indicating the degree of dispersion of the normalized response value within the current stage; A statistic representing the dispersion of the normalized response values ​​within the reference phase; This indicates the structural partition correction amount for region identifier r; and These represent the mean statistic and the dispersion statistic of the normalized response values ​​within the structural partition where region identifier r is located, respectively. and These represent the global mean statistic and the global dispersion statistic of the whole wall's normalized response value, respectively. , , and These represent the correction weights. By constructing these correction amounts, the risk of threshold shift caused by stage changes and structural non-uniformity can be explicitly compensated.

[0100] After obtaining the stage drift correction and structural partition correction, the initial global baseline threshold and initial local baseline threshold are finalized by combining these two values ​​to form the global baseline threshold and local baseline threshold. In this embodiment, the final correction process refers to further compensating the initial threshold, obtained solely based on background statistical features, by incorporating the stage state and structural partition state in the current detection scenario. This ensures that the final threshold possesses global uniformity, local adaptability, and scene consistency. Here, the global baseline threshold refers to the final background threshold used for global anomaly detection across the entire wall surface; the local baseline threshold refers to the final background threshold used for local anomaly detection around each area marker. Its expression is:

[0101] in, Indicates the global baseline threshold; This represents the local baseline threshold for the region identified as r; Indicates the initial global baseline threshold; This represents the initial local baseline threshold for the region identified as r; Indicates the stage drift correction amount; This represents the structural partition correction amount for region identifier r. This final correction process enables subsequent threshold mapping and region extraction to more accurately distinguish between background regions and actual defect regions in low-temperature cyclic building wall scenarios, and provides a reliable threshold basis for the stable generation of defect region sets.

[0102] In one possible implementation, threshold mapping and region extraction are performed based on a stable defect response distribution field to generate a set of defect regions and output the internal defect identification result. Specifically, this includes: performing response value normalization on the stable defect response distribution field to form a normalized response field; performing dual threshold mapping by combining a global baseline threshold and a local baseline threshold to form an initial defect candidate label field; performing connectivity expansion based on the initial defect candidate label field to form a set of candidate defect connected domains; performing geometric consistency screening based on the set of candidate defect connected domains to form a set of defect regions; performing defect depth estimation based on the set of defect regions to form a defect depth parameter; and performing defect severity grading based on the defect depth parameter, the normalized response field, and the set of candidate defect connected domains to obtain the internal defect identification result.

[0103] Specifically, the stable defect response distribution field is first subjected to response value normalization to form a normalized response field. In this embodiment, response value normalization refers to compressing the stable defect response values ​​corresponding to each region identifier in the stable defect response distribution field to a unified relative response interval, making the response values ​​under different wall types, different detection rounds, and different coupling discrimination strengths comparable. Specifically, the minimum and maximum response values ​​in the current stable defect response distribution field are first calculated, and then the stable defect response values ​​of each region identifier are linearly normalized to obtain the normalized response field. Here, the normalized response field refers to a response field that expresses the relative anomaly degree of each region at a unified scale, which can be directly used for subsequent threshold mapping and region extraction. The expression for the normalized response value is:

[0104] in, This represents the normalized response value for the region identifier r, and its value is between zero and one. This represents the stable defect response value for region identifier r; and These represent the minimum and maximum values ​​in the current stable defect response distribution field, respectively. This indicates a stable term. The purpose of this processing is to transform absolute response differences into relative anomaly intensities, facilitating standardized threshold judgment.

[0105] After obtaining the normalized response field, a dual-threshold mapping is performed by combining the global baseline threshold and the local baseline threshold to form an initial defect candidate labeling field. In this embodiment, dual-threshold mapping refers to simultaneously using the global baseline threshold and the local baseline threshold to discriminate the normalized response field, thus taking into account both overall background consistency and local background differences. Specifically, the global baseline threshold and the corresponding local baseline threshold for each region are first mapped to the normalized response space, and then strong threshold determination and weak threshold determination are performed on each region identifier. Here, the initial defect candidate labeling field refers to the spatial distribution field recording the binary labeling results of each region identifier under strong threshold conditions and weak threshold conditions. The expressions for strong threshold and weak threshold determination are:

[0106] in, The strong threshold marker for the region identifier r has a value of one, indicating that the region meets the strong anomaly condition. This represents a weak threshold marker for the region identified as r, where a value of one indicates that the region meets the weak anomaly condition. Indicates the global baseline threshold; This represents the local baseline threshold, denoted as r, for the region. Through dual-threshold mapping, the core defect region and its peripheral regions can be identified.

[0107] After forming the initial candidate defect labeling field, connectivity expansion is performed based on the initial candidate defect labeling field to form a set of candidate defect connected regions. In this embodiment, connectivity expansion refers to using a strong threshold labeled region as the core seed, and gradually expanding and merging regions that are spatially connected to it and satisfy the weak threshold condition, thereby forming a continuous defect region. Specifically, for each region with a strong threshold label of one, its neighborhood set is traversed, and all regions with a weak threshold label of one and connected to it are included in the same connected region, until no further expansion is possible. Here, the set of candidate defect connected regions refers to a set composed of multiple spatially connected regions, with each connected region corresponding to a potential defect region. To quantify connectivity, a connectivity support coefficient can be defined:

[0108] in, This represents the connectivity support coefficient for region denoted as r. The larger the value, the more weak threshold support regions exist around this region. This represents the set of neighborhoods identified by region r. Through connectivity expansion, scattered weak anomaly regions and core anomaly regions can be unified and integrated into a complete defect region.

[0109] After obtaining the set of candidate defect connected components, geometric consistency screening is performed based on this set to form a defect region set. In this embodiment, geometric consistency screening refers to analyzing the spatial morphological characteristics of connected components to eliminate irregular regions formed by noise or boundary perturbations, retaining only connected components that conform to the actual defect morphology. Specifically, the area, boundary length, and compactness of each candidate connected component are calculated, and screening is performed according to preset geometric constraints. The expression for the compactness parameter is:

[0110] in, The compactness parameter represents the m-th connected component; a larger value indicates a more regular shape. Represents the area of ​​a connected region; This represents the length of the connected domain boundary. Geometric consistency filtering can effectively eliminate scattered noise regions and abnormal boundary regions, resulting in a set of stable defect regions.

[0111] After forming a set of defect regions, defect depth estimation is performed based on this set to generate defect depth parameters. In this embodiment, defect depth estimation refers to indirectly estimating the longitudinal location or severity of defects within the wall using thermal response delay characteristics and acoustic propagation disturbance characteristics. Specifically, for each defect region, the mean values ​​of its temperature recovery delay characterization, acoustic propagation disturbance characterization, and normalized response are statistically analyzed, and depth parameters are constructed by combining these with the region's morphological characteristics. The expression for the defect depth parameters is:

[0112] in, The parameter representing the defect depth of the m-th defect region; This represents the mean of the characteristic quantity of temperature recovery delay in this region; This represents the mean value of the quantity characterizing acoustic propagation disturbance. This represents the mean of the normalized response values; This indicates the tightness parameter; , , and This indicates the weight. This parameter reflects the depth of the defect and the degree of its impact on the structure.

[0113] After obtaining the defect depth parameters, a defect severity grading process is performed based on the defect depth parameters, the normalized response field, and the set of candidate defect connected components to obtain the internal defect identification results. In this embodiment, the defect severity grading process refers to classifying defect regions by comprehensively considering anomaly intensity, spatial range, and depth risk. Specifically, a severity index is constructed for each defect region, and it is classified according to a preset grading interval. The expression for the severity index is:

[0114] in, This represents the severity index of the m-th defect region; This represents the mean of the normalized response; Indicates the area of ​​the region; Indicates the defect depth parameter; This represents the mean of the connectivity support coefficients; , , and The weights are represented by the values. Finally, the spatial location, area, depth parameters, and severity level of each defect region are output, thus forming the internal defect identification result.

[0115] This embodiment also discloses a device for analyzing internal defects in building walls that integrates infrared thermal imaging and acoustic data, referring to... Figure 3 The device includes an acquisition module 301, a processing module 302, and an output module 303. It is used to execute the aforementioned method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data, wherein: The acquisition module 301 is used to acquire the infrared thermal image sequence and acoustic response sequence corresponding to the target wall and perform spatial registration processing to form a coupled observation data set. Processing module 302 is used to construct a temperature recovery delay characterization quantity and a sound propagation disturbance characterization quantity based on the coupled observation data set to form a defect response feature set; Processing module 302 is used to construct a coupled discriminant function for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response; Processing module 302 is used to perform spatial continuity constraint processing and stage consistency correction processing according to the coupling discriminant function to obtain a stable defect response distribution field; The output module 303 is used to perform threshold mapping and region extraction processing based on a stable defect response distribution field, generate a set of defect regions, and output the internal defect identification results.

[0116] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0117] This embodiment also discloses an electronic device, as shown in the reference. Figure 4The electronic device may include: at least one processor 401, at least one communication bus 402, user interface 403, network interface 404, and at least one memory 405.

[0118] The communication bus 402 is used to enable communication between these components.

[0119] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0120] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0121] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0122] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. As a computer storage medium, the memory 405 may include an operating system, a network communication module, a user interface 403 module, and an application program for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data.

[0123] exist Figure 4 In the electronic device shown, the user interface 403 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 401 can be used to call the application program stored in the memory 405 for analyzing the internal defects of building walls by fusing infrared thermal imaging and acoustic wave data. When executed by one or more processors 401, the electronic device performs one or more methods as described in the above embodiments.

[0124] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0125] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0126] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0129] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 405 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned memory 405 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0130] The present invention also discloses a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors 401, these instructions cause an electronic device to perform one or more methods as described in the above embodiments.

[0131] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truths. This invention is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for analyzing internal defects in building walls by integrating infrared thermal imaging and acoustic wave data, characterized in that, The method includes: The infrared thermal image sequence and acoustic response sequence corresponding to the target wall are acquired and spatial registration processing is performed to form a coupled observation data set. Based on the aforementioned coupled observation data set, temperature recovery delay characterization and acoustic propagation disturbance characterization are constructed to form a defect response feature set; A coupled discriminant function is constructed for the set of defect response features to achieve fusion characterization of infrared thermography and acoustic response; Based on the coupling discriminant function, spatial continuity constraint processing and stage consistency correction processing are performed to obtain a stable defect response distribution field; Based on the stable defect response distribution field, threshold mapping and region extraction processing are performed to generate a set of defect regions and output the internal defect identification results.

2. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 1, characterized in that, The construction of a temperature recovery delay characterization quantity and a sound propagation disturbance characterization quantity based on the coupled observation data set to form a defect response feature set specifically includes: The coupled observation data set is subjected to observation unit partitioning to obtain a regional observation mapping set; Based on the regional observation mapping set, thermal image benchmark straightening processing is performed on the infrared thermal image sequence to obtain a regional thermal response trajectory set and a reference thermal response trajectory. A temperature recovery delay characterization quantity is constructed based on the set of regional thermal response trajectories and the reference thermal response trajectory. Based on the regional observation mapping set, propagation path analysis and path perturbation correction are performed on the acoustic response sequence to form a regional acoustic response trajectory set and a reference acoustic response trajectory. A sound propagation disturbance characterization quantity is constructed based on the set of regional acoustic response trajectories and the reference acoustic response trajectory; Based on the temperature recovery delay characterization and the acoustic propagation disturbance characterization, a screening process is performed to form a candidate set of regional defects. The temperature recovery delay characterization, acoustic propagation disturbance characterization, and auxiliary constraint characterization corresponding to each region identifier in the candidate set of regional defects are bound together to form the defect response feature set.

3. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 2, characterized in that, The step of constructing a temperature recovery delay characterization quantity based on the set of regional thermal response trajectories and the reference thermal response trajectory specifically includes: Recovery phase identification is performed on the set of regional thermal response trajectories and the reference thermal response trajectory to form a set of regional recovery time periods; Temperature trajectory realignment is performed on the regional thermal response trajectory and the reference thermal response trajectory using the set of regional recovery time periods to obtain the realigned recovery curve; Based on the normalized recovery curve, extract the regional recovery node sequence and the reference recovery node sequence corresponding to the recovery ratio node; Construct node delay components based on the region recovery node sequence and the reference recovery node sequence; Based on the normalized recovery curve, a trajectory offset component and a slope hysteresis component are constructed. The temperature recovery delay characterization quantity is constructed based on the node delay component, the trajectory offset component, and the slope hysteresis component.

4. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 1, characterized in that, The construction of a coupled discriminant function for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response specifically includes: Thermal deviation component, acoustic deviation component and constraint deviation vector are constructed based on the region-corrected feature set, wherein the region-corrected feature set is formed by performing feature correction processing on the defect response feature set; Based on the thermal deviation component and the acoustic deviation component, a thermo-acoustic consistent coupling component and a thermo-acoustic difference suppression component are constructed, and a region confidence coefficient is constructed based on the constraint deviation vector. Based on the thermal deviation component, the acoustic deviation component, the thermo-acoustic consistent coupling component, the thermo-acoustic difference suppression component, and the region confidence coefficient, a coupling discriminant function is constructed to obtain the coupling discriminant value; The coupling discriminant value is subjected to time-series accumulation processing based on the consecutive observation rounds to form a time-series coupling discriminant value; Based on the temporal coupling discriminant value, perform neighborhood collaborative enhancement processing to form a region fusion characterization value; Based on the region fusion characterization value, a discriminative interval mapping process is performed to form the fusion response probability.

5. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 1, characterized in that, The step of performing spatial continuity constraint processing and stage consistency correction processing based on the coupling discriminant function to obtain a stable defect response distribution field specifically includes: An initial discrimination distribution field is generated for each region identifier based on the aforementioned coupled discrimination function; Construct a regional neighborhood weight set based on the initial discrimination distribution field; Spatial continuity constraint processing is performed on the initial discrimination distribution field using the regional neighborhood weight set to obtain spatial constraint discrimination values; Based on the spatial constraint discrimination value, perform boundary preservation enhancement to obtain the boundary preservation discrimination value; Based on the boundary-preserving discriminant values, stage division is performed to form a stage division set and a stage discriminant sequence is constructed; A stage consistency coefficient is constructed based on the stage discrimination sequence; Based on the stage consistency coefficient, the boundary preservation discriminant value is subjected to stage consistency correction processing to obtain the stage correction discriminant value; Based on the stage correction discriminant value, perform multi-round cumulative fusion processing to obtain stable defect response values ​​and map them to the stable defect response distribution field.

6. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 1, characterized in that, Before performing threshold mapping and region extraction processing based on the stable defect response distribution field to generate a defect region set and output the internal defect identification result, the method further includes: Background sample screening is performed based on the normalized response field to obtain a set of background candidate regions, wherein the normalized response field is obtained by performing response value normalization processing on the stable defect response distribution field; Global background statistical modeling and global tail risk estimation are performed based on the set of background candidate regions to form global background center parameters, global background discrete parameters, and global tail boundary parameters. Global threshold fusion processing is performed based on the global background center parameter, the global background discrete parameter, and the global tail boundary parameter to form an initial global baseline threshold. Based on the normalized response field, local neighborhood modeling and local background denoising are performed around each region identifier to form a set of local background candidate samples; Local robust statistical modeling is performed on the local background candidate sample set to form local background center parameters, local background discrete parameters, and local tail boundary parameters; Local threshold fusion is performed based on the local background center parameter, the local background discrete parameter, and the local tail boundary parameter, combined with the local complexity coefficient, to form an initial local baseline threshold. Based on the normalized response field, a correction process is performed to form a stage drift correction amount and a structural partitioning correction amount; By combining the stage drift correction amount and the structural partition correction amount, a final correction process is performed on the initial global baseline threshold and the initial local baseline threshold to form the global baseline threshold and the local baseline threshold.

7. The method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic wave data according to claim 6, characterized in that, The process of performing threshold mapping and region extraction based on the stable defect response distribution field to generate a set of defect regions and output the internal defect identification results specifically includes: The stable defect response distribution field is subjected to response value normalization processing to form a normalized response field; A dual-threshold mapping is performed by combining the global baseline threshold and the local baseline threshold to form an initial defect candidate label field; Based on the initial defect candidate label field, perform connectivity expansion to form a set of candidate defect connectivity domains; Geometric consistency screening is performed based on the candidate defect connected component set to form a defect region set; Defect depth estimation is performed based on the set of defect regions to form defect depth parameters; Based on the defect depth parameter, the normalized response field, and the set of candidate defect connected domains, defect severity classification processing is performed to obtain the internal defect identification result.

8. A device for analyzing internal defects in building walls that integrates infrared thermal imaging and acoustic wave data, characterized in that, The device is used to perform the method for analyzing internal defects in building walls by fusing infrared thermal imaging and acoustic data as described in any one of claims 1-7. The device includes an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire the infrared thermal image sequence and acoustic response sequence corresponding to the target wall and perform spatial registration processing to form a coupled observation data set. The processing module is used to construct a temperature recovery delay characterization quantity and a sound propagation disturbance characterization quantity based on the coupled observation data set to form a defect response feature set. The processing module is used to construct a coupled discriminant function for the defect response feature set to achieve fusion characterization of infrared thermography and acoustic response; The processing module is used to perform spatial continuity constraint processing and stage consistency correction processing according to the coupling discriminant function to obtain a stable defect response distribution field; The output module is used to perform threshold mapping and region extraction processing based on the stable defect response distribution field, generate a set of defect regions, and output the internal defect identification results.

9. An electronic device, characterized in that, The device includes a processor, a communication bus, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The communication bus is used to enable communication between the components within the electronic device. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images