Method and system for enhancing features of hollowing defects of external wall based on time-lapse thermography

Abstract

The application provides a delay thermal imaging-based feature enhancement construction method and system for external wall hollowing defects, relates to the cross technical fields of building engineering nondestructive testing, infrared thermal imaging application and digital signal processing, and comprises the following steps: determining optimal dynamic detection time windows and sampling interval length parameters according to the wall surface orientation of a target building external wall and the daily cycle evolution law of the surface temperature field; acquiring a continuous three-dimensional thermal image sequence of the target building external wall; according to the start and stop instructions of the dynamic detection time windows, intercepting a short-time continuous thermal response sequence with a corresponding length from the continuous three-dimensional thermal image sequence; performing dimension reduction, transformation and feature enhancement processing on the intercepted short-time continuous thermal response sequence based on a partial least square method, and constructing a feature image representing the external wall hollowing defects; and the application breaks through the limitation of complex natural environment interference, and realizes absolute separation and high-precision enhancement extraction of weak hollowing defects.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of non-destructive testing in building engineering, infrared thermal imaging applications, and digital signal processing, specifically to a method and system for enhancing the features of external wall hollow defects based on time-delay thermal imaging. Background Technology

[0002] In recent years, the application of external wall insulation systems (such as EPS molded polystyrene facing brick structures) in the operation and maintenance phase of existing buildings has become increasingly widespread. Their durability and safety have become core pain points in building safety management. Exterior walls, constantly exposed to harsh natural environments, are highly susceptible to hollowing defects due to the failure of the internal bonding of the adhesive mortar layer caused by continuous thermal expansion and contraction from solar radiation, freeze-thaw cycles from rainwater infiltration, and the material's own creep and aging. If such hidden dangers are not detected in time, the accumulated damage will lead to large-scale detachment of facing bricks, not only damaging the building's original waterproofing and insulation functions but also posing a serious threat to public safety. Against this backdrop, how to achieve real-time, non-destructive, and high-precision monitoring of external wall hollowing defects has become a key technical problem that the industry urgently needs to solve.

[0003] Traditional methods for detecting hollow areas in exterior walls primarily rely on physical tapping, judging the quality by manually identifying changes in the frequency of the tapping sound. However, this method requires the installation of high-altitude equipment, resulting in low efficiency, high cost, and a heavy reliance on the operator's subjective auditory experience, making it difficult to generate objective and quantifiable digital records. Subsequent developments in infrared single-frame thermal imaging technology, while utilizing the difference in thermal resistance between hollow defects (air gaps) and normal bonding mortar to attempt to identify defects through surface temperature anomalies under natural heat flow excitation, face significant challenges in real-world environments. These include periodic variations in solar radiation, dynamic obstruction by clouds and shadows, irregular wind speed fluctuations, uneven emissivity of the facing brick surface, and electronic noise from the detector itself. These random interferences easily drown out the weak defect temperature difference signal due to significant background temperature drift and spatial noise, leading to blurred defect boundaries and extremely low contrast. This severely limits the reliability and engineering applicability of this technology.

[0004] Therefore, existing technologies suffer from the problem that single-frame heatmaps are easily overwhelmed by strong noise, resulting in low defect identification, and lack effective means to extract weak feature signals from complex time sequences, which seriously affects the detection accuracy and stability of practical engineering applications. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method and system for enhancing the features of external wall hollow defects based on time-delay thermal imaging. This method overcomes the limitations imposed by complex natural environmental interference and achieves absolute separation and high-precision enhanced extraction of subtle hollow defects.

[0006] According to some embodiments, the present invention adopts the following technical solution: A method for enhancing the features of external wall hollow defects based on time-lapse thermal imaging includes: Based on the diurnal evolution of the wall orientation and surface temperature field of the target building's exterior wall, the optimal dynamic detection time window and sampling interval parameters are determined. A continuous three-dimensional thermal image sequence of the exterior wall of the target building is obtained. The sequence is based on time-lapse thermal imaging and is obtained by taking continuous images of the exterior wall of the target building at equal time intervals using a sampling interval parameter. Based on the start and end instructions of the dynamic detection time window, a short-time continuous thermal response sequence of corresponding length is extracted from the continuous three-dimensional thermal image sequence; Based on partial least squares, the short-time continuous thermal response sequence is subjected to dimensionality reduction, transformation and feature enhancement to construct a feature image characterizing the hollow defects of the exterior wall.

[0007] According to some embodiments, the present invention adopts the following technical solution: A system for enhancing the features of external wall hollow defects based on time-lapse thermal imaging includes: The determination module is configured to: determine the optimal dynamic detection time window and sampling interval parameters based on the diurnal evolution of the wall orientation and surface temperature field of the target building's exterior wall; The acquisition module is configured to acquire a continuous three-dimensional thermal image sequence of the exterior wall of the target building, wherein the sequence is obtained by taking continuous images of the exterior wall of the target building at equal time intervals based on time-lapse thermal imaging and using a sampling interval duration parameter; The interception module is configured to: intercept a short-time continuous thermal response sequence of corresponding length from the continuous three-dimensional thermal image sequence according to the start and end instructions of the dynamic detection time window; The construction module is configured to: perform dimensionality reduction, transformation and feature enhancement processing on the extracted short-time continuous thermal response sequence based on partial least squares method to construct a feature image characterizing the hollow defects of the external wall.

[0008] According to some embodiments, the present invention adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the method for enhancing the construction of external wall hollow defect features based on time-delayed thermal imaging.

[0009] According to some embodiments, the present invention adopts the following technical solution: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for enhancing the construction of external wall void defect features based on time-delayed thermal imaging.

[0010] According to some embodiments, the present invention adopts the following technical solution: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for enhancing the features of external wall voids based on time-delay thermal imaging.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention breaks away from the conventional thinking that "a complete 24-hour day-night cycle must be recorded for time-series analysis" in traditional research. It innovatively proposes a dynamic detection time window based on the "extreme value neighborhood scheme." By accurately locking the extreme moment of the external wall temperature and extracting a small sequence before and after it, it can significantly shorten the working time of on-site engineers and the time occupied by equipment, while perfectly preserving the most discriminative dynamic thermal response characteristics, greatly improving detection efficiency. It abandons the traditional single-image smoothing technique (which often blurs the physical boundaries of defects) and creatively introduces the partial least squares (PLS) method from the field of signal processing into building thermal engineering. It can directly extract the first-order latent variable representing the thermal resistance anomaly of local defects from the time-domain matrix based on the mathematical differences in temperature evolution. This mathematical dimensionality reduction extraction allows the method to effectively ignore high-frequency random natural noise such as wind speed fluctuations and cloud cover, ensuring high detection results. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0013] Figure 1 This is a flowchart of the method for enhancing the features of external wall hollow defects based on time-delayed thermal imaging in Example 1. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0017] Example 1 One embodiment of the present invention provides a method for enhancing the construction of external wall hollow defect features based on time-delayed thermal imaging, such as... Figure 1 As shown, it includes: Step S1: Determine the optimal dynamic detection time window and sampling interval parameters based on the wall orientation and diurnal evolution of the surface temperature field of the target building's exterior wall; Step S2: Obtain a continuous three-dimensional thermal image sequence of the target building's exterior wall. The sequence is obtained by taking continuous photos of the target building's exterior wall at equal time intervals based on time-lapse thermal imaging and using a sampling interval parameter. Step S3: Based on the start and end instructions of the dynamic detection time window, extract a short-time continuous thermal response sequence of corresponding length from the continuous three-dimensional thermal image sequence; Step S4: Based on the partial least squares method, the extracted short-time continuous thermal response sequence is subjected to dimensionality reduction, transformation and feature enhancement processing to construct a feature image characterizing the hollow defects of the external wall.

[0018] As one embodiment, the method for enhancing and constructing features of external wall hollow defects based on time-delay thermal imaging of the present invention overcomes the limitations of interference from complex natural environments, and achieves absolute separation and high-precision enhanced extraction of weak hollow defects. The specific implementation process is described below from the parameter determination step, data acquisition step, sequence truncation step, and feature construction step: I. Parameter determination steps, used to determine a strictly matching dynamic detection time window and optimized sampling interval duration based on the wall orientation (east, south, west, north) and the diurnal evolution law of surface heat flow of the actual building exterior wall; This step aims to address the questions of "when to take pictures" and "how often to take pictures" in traditional methods. By accurately matching the diurnal thermal evolution pattern of the wall's orientation, it minimizes data collection and on-site operation time while ensuring defect identifiability. Therefore, the core lies in differentiating the "dynamic detection time window" and "sampling interval" based on the thermophysical characteristics of the wall's orientation. Specifically: 1. Determining the dynamic detection time window (extreme value neighborhood window) The core principle is that the temperature difference caused by hollow defects is most significant when the wall surface temperature changes at its highest rate (rapid heating or cooling). Therefore, instead of 24-hour monitoring, the system focuses on the moment when the surface temperature reaches its extreme value (highest or lowest), and uses the immediate vicinity before and after this point as the detection window. Specifically: First, determine the orientation by using an electronic compass or architectural drawings to determine the orientation (east, south, west, or north) of the exterior wall to be measured.

[0019] Then, extreme value time prediction is performed. Based on local meteorological parameters (or short-term test curves on site), the extreme value time when the surface temperature of the wall facing that direction reaches the highest point (sunny side) or the lowest point (shaded side) of the day is estimated. For example, the east wall is heated by morning sunlight, and the extreme value is about 09:00. The south wall is heated by strong midday sunlight, and the extreme value is about 13:00. The west wall is heated by afternoon afternoon sunlight, and the extreme value is about 16:00. The north wall has no strong direct sunlight throughout the day, and the thermal excitation is very weak. There is no significant extreme value of temperature rise. It is necessary to rely on long-term accumulation or degradation to detect the lowest temperature time (about 14:30).

[0020] Finally, the window span is set (including engineering tolerances). For the sunny walls (east, south, and west), a 3-hour or 4-hour window with extreme value neighborhood is used, extending equidistantly before and after the extreme value moment. Considering the on-site half-hour operation tolerance, the hard cut-off standard is as follows: The east wall section was captured from 07:15 to 10:45 (the extreme moment was 09:00, with a tolerance of approximately 1.5 hours before and after).

[0021] The south wall section is captured from 11:15 to 14:45 (the extreme moment is 13:00).

[0022] The west wall section is captured from 14:15 to 17:45 (the extreme moment is 16:00).

[0023] For the shaded wall (north-facing wall), one of the two options will be used: Option 1 (Recommended): Use a complete 24-hour observation cycle.

[0024] Option 2 (Degeneracy): Use an extreme value neighborhood 4-hour window to capture 13:15 ~ 17:45 (due to the lack of strong stimulus, a longer window is needed to accumulate weak signals).

[0025] 2. Optimization of the sampling interval parameter The core principle is to maximize data compression while ensuring sufficient temporal resolution to characterize temperature evolution, thus adapting to the bandwidth and battery life of inspection equipment such as drones. Specifically, the shutter trigger frequency of the infrared camera is dynamically adjusted according to the wall orientation, including: (1) Sunlit walls (east, south, west): The sampling interval is set to any integer value less than 30 minutes (e.g., 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes).

[0026] The preferred value, recommended in engineering practice, is 10 minutes; because extending from 5 minutes to 10 minutes results in a very small decrease in signal-to-noise ratio (<5%), but halves the amount of data, making it extremely efficient; however, the interval should not exceed 30 minutes, otherwise it will be impossible to capture enough thermal excitation evolution nodes.

[0027] (2) Shaded wall (north wall): The sampling interval is set to any integer value not exceeding 10 minutes (e.g., 1 minute, 2 minutes, 5 minutes, 10 minutes).

[0028] To apply this method on a large scale to UAV swarm inspection, the data transmission volume must be compressed, i.e., the "sampling interval" of the images must be optimized. In this embodiment, based on the extreme neighborhood of 3 hours and 4 hours, exhaustive attenuation tests were conducted on sampling intervals of 5, 10, 30, 60, 90, and 120 minutes. The data shows that when the sampling interval is extended from 5 minutes to 10 minutes, the decrease in signal-to-noise ratio when processing extreme neighborhood segments is minimal and well within the engineering tolerance. When the interval is widened to 30 minutes, sufficient node information to characterize the thermal excitation evolution can still be retained for sun-exposed walls under strong sunlight, and the algorithm can still separate effective features. However, once the sampling interval reaches or exceeds 60 minutes, the defect enhancement capability of all algorithms is lost. Therefore, the preferred value is 5 minutes. Because the thermal excitation of the north wall is extremely weak and the temperature change signal amplitude is small, a higher time sampling rate is needed to extract the weak defect feature changes from the noise.

[0029] 3. Final Output This step outputs the following two key parameter sets for use in subsequent steps (data acquisition and sequence truncation): Dynamic detection time window: clearly defined start and end timestamps (e.g., East Wall: 07:15-10:45).

[0030] Sampling interval duration: fixed time interval (e.g., sunny side: 10 minutes; shady side: 5 minutes).

[0031] II. Data acquisition steps are used to build a hardware detection platform. The infrared thermal imager of the high-precision focal plane array detector is integrated into the UAV to acquire the natural time-series thermal response data of the target building's exterior wall under the excitation of natural environmental heat flow, forming a natural, noisy, continuous three-dimensional thermal image sequence (spatial dimension X, spatial dimension Y, and time dimension T). The core of this step lies in using drones as mobile platforms to achieve efficient inspections of high-rise buildings, large building complexes, and hard-to-reach areas; simultaneously, the collected data retains its "naturally noisy" nature, without any artificial filtering or noise reduction preprocessing, thus fully preserving information about complex environmental interference. Specifically: 1. Hardware Platform Construction The drone platform uses an industrial-grade multi-rotor drone, featuring RTK high-precision positioning, a flight time of at least 30 minutes, and wind resistance up to level 5. An infrared thermal imager is mounted below the drone via a three-axis stabilized gimbal, with a thermal sensitivity better than 0.05 degrees Celsius and a resolution of at least 640×512 pixels.

[0032] 2. Data Collection and Execution The drone automatically takes off and flies to the first waypoint along a preset route. Upon arrival, it switches to hover mode, adjusts the gimbal to make the thermal imager's optical axis perpendicular to the wall, and automatically triggers shooting according to the optimized sampling interval (e.g., 10 minutes on the sunny side and 5 minutes on the shaded side). Each frame of raw thermal image data is stored in real time to the onboard memory.

[0033] After collecting data at a single waypoint, the drone automatically flies to the next waypoint and repeats the hovering data collection process until the entire detection area is covered.

[0034] 3. Data Output After the acquisition is completed, all the acquired two-dimensional thermal images are organized into a continuous three-dimensional thermal image sequence in chronological order. The data structure of this sequence is a three-dimensional tensor consisting of the height pixel count multiplied by the width pixel count multiplied by the total number of sampling times. The height and width dimensions represent spatial locations and together describe the spatial temperature distribution on the exterior wall surface. The time dimension represents the sampling timestamp and describes the temperature evolution trajectory of each spatial location over time.

[0035] Simultaneously, the acquired parameters, flight trajectory, and environmental information are encapsulated, and this naturally noisy sequence is passed as output to the subsequent sequence truncation step.

[0036] The third step is a sequence truncation step, which is used to accurately truncate a short-time continuous thermal response sequence of the required length from the continuous three-dimensional thermal image sequence with the long time span within the dynamic detection time window. This step is used to accurately extract a short-time continuous thermal response sequence of the required length from the long-span continuous three-dimensional thermal image sequence obtained in the data acquisition step within the dynamic detection time window. Its core is to reduce the original massive data of the whole time period to only the time segment containing the most discriminative information, so as to provide compact and information-rich data input for subsequent signal processing steps.

[0037] The input to the sequence extraction step is a continuous three-dimensional thermal image sequence output from the data acquisition step. The data structure of this sequence is a three-dimensional tensor consisting of the height in pixels multiplied by the width in pixels multiplied by the total number of sampling times, where the time dimension covers the entire acquisition duration. For a sunlit wall, the original sequence may contain dozens of thermal images spanning 3.5 hours; for a north-facing wall observed in all weather conditions, the original sequence may contain hundreds of thermal images spanning 24 hours. The sequence extraction step needs to select a subset from these original frames.

[0038] The selection is based on the dynamic detection time window output by the parameter determination step. This window defines the start and end boundaries of the data extraction. The specific extraction process is as follows: 1. Start Frame Positioning Traverse the timestamps of the continuous 3D thermal image sequence, find the first thermal image frame that is greater than or equal to the window start time, and mark its index as the start frame index; if the window start time falls between two frames (i.e. there is no exact matching timestamp), select the first frame with a timestamp greater than the window start time as the start frame, or select the frame with the closest time using the nearest neighbor principle.

[0039] 2. End frame positioning Similarly, iterate through the timestamp sequence to find the last thermal image frame whose time is less than or equal to the window end time, and mark its index as the end frame index; if the window end time falls between two frames, select the last frame whose timestamp is less than the window end time as the end frame.

[0040] 3. Subsequence Extraction Based on the start and end frame indices, continuous subsequences are extracted from the original 3D sequence along the time dimension to form a short-time continuous thermal response sequence; the data structure of this sequence is highly... Multiply by width Multiply by the number of frames captured The three-dimensional tensor is represented by each frame, which represents the temperature field distribution of the outer wall surface at a certain sampling moment within the detection time window. The entire sequence records the temperature evolution trajectory of each pixel within the window over time. The sequence retains its natural noisy properties and has not undergone any pre-filtering or smoothing processing.

[0041] IV. Feature Construction Steps: The improved partial least squares method is used to perform dimensionality reduction, transformation, and feature enhancement processing on the extracted natural short-time continuous thermal response sequence to construct a feature image for accurately characterizing the hollow defects of the exterior wall.

[0042] The partial least squares method is effective in constructing features because the temperature change in the hollow area always lags behind or differs from the overall temperature change of the normal wall surface due to air insulation. The partial least squares method is good at capturing this synchronicity difference. To this end, the background temperature change and local defect anomaly in the short-time continuous thermal response sequence are separated. This step decouples the time trajectory of temperature change from the overall thermal excitation environment by regression. The partial least squares method is introduced from the traditional field of chemometrics into building thermal engineering. By establishing the regression relationship between temperature evolution and background environmental variables, the feature image corresponding to the first-order latent variable representing the local thermal resistance anomaly is extracted from the naturally noisy time series data.

[0043] Traditional partial least squares method considers both the independent variable matrix and the independent variable matrix. and dependent variable matrix The covariance structure is used to establish a regression model between the two by extracting latent variables; this embodiment makes targeted improvements to the traditional method: the short-time continuous thermal response sequence is used as the input matrix after dynamic background subtraction. Based on the overall average temperature curve of the exterior wall within the detection time window and its first and second derivatives, a response matrix is ​​constructed. Then, the spatial weight vectors corresponding to the first-order latent variables are extracted and reconstructed into a two-dimensional feature image.

[0044] The improvements in this embodiment are reflected in three aspects: First, dynamic background subtraction preprocessing is introduced to subtract the global average temperature at each time frame segment, thereby weakening the dominant role of background temperature drift; Second, based on the overall average temperature curve, its first and second derivatives are introduced into the response matrix to characterize the temperature change rate and transition features, respectively; Third, instead of reconstructing the thermal image sequence through latent variables, the two-dimensional feature map is directly recovered using the spatial weight vector corresponding to the first latent variable.

[0045] Provide a concrete example of feature construction, including: 1. Dynamic background deduction Because ambient temperature (such as sunlight) causes the overall temperature of the wall to rise, this global change can overwhelm local defect signals. Therefore, it is necessary to remove this background first, that is, to perform dynamic background removal processing on the extracted short-time continuous thermal response sequence, specifically: First, iterate through each time frame in the sequence and calculate the average temperature of all spatial pixels within that frame to obtain the overall average temperature of the exterior wall surface at that moment. Then, subtract this overall average temperature from the temperature value of each pixel within that frame to obtain the residual matrix after background subtraction for that frame, expressed by the formula:

[0046] in, for The spatial temperature distribution on the exterior wall surface at any given time. This represents the average temperature of all pixels on the exterior wall surface at that moment.

[0047] Repeat the above operation for all time frames to obtain a new three-dimensional sequence, where the value of each pixel represents the offset of that position relative to the global average temperature. The core function of this process is to eliminate the global background temperature drift caused by the overall rise and fall of the ambient temperature, so that subsequent analysis can focus on the abnormal thermal response of local areas.

[0048] 2. Construct the input matrix

[0049] The three-dimensional sequence obtained after dynamic background subtraction is reconstructed into a two-dimensional matrix, which serves as the input matrix for partial least squares. The specific reconstruction method is as follows: First, flatten the spatial dimension of each frame image into a one-dimensional vector with a length of Multiply Then, arrange the flattened vectors of all time frames column-wise to form a vector with dimension ( Multiply by The matrix consists of columns, where each row represents the temperature offset response curve of a spatial pixel at all sampling times, and each column represents the spatial distribution of all pixels at a given time.

[0050] 3. Construct the response matrix

[0051] Constructing the response matrix using partial least squares The specific reconstruction method is as follows: First, the overall average temperature value of the exterior wall surface at each sampling moment in the extracted short-time continuous thermal response sequence is extracted to form an overall average temperature curve that varies with time. Then, the first derivative of this curve is calculated to obtain the temperature change rate curve, reflecting how quickly the wall surface heats up or cools down. Next, the second derivative of this curve is calculated to obtain the temperature change acceleration curve, characterizing the turning point features in the temperature evolution process.

[0052] The two curves obtained from the above differentiation are concatenated row by row to form a curve with a dimension of 2 rows multiplied by 1. Column response matrix This matrix represents the dynamic change pattern of the overall environmental heat flux within the detection window, expressed by the formula:

[0053] Wherein, the first derivative The second derivative reflects the rate of temperature change. It is used to depict the transitional features in the process of temperature change.

[0054] 4. Partial Least Squares Decomposition The input for this step is the thermal image sequence matrix after dynamic background subtraction. and thermal process reference response matrix , respectively represented as:

[0055]

[0056] In the formula, To detect the number of thermal images within a time window, This represents the total number of pixels in a single frame of a thermal image.

[0057] The goal of partial least squares decomposition is to optimize the thermal image sequence matrix. and response matrix Find a pair of latent variables between them such that their covariance in the latent variable space is maximized. For the th... The score vector corresponding to the thermal image sequence matrix of the 1 latent variables can be represented as:

[0058] The score vector corresponding to the response matrix can be represented as:

[0059] in, and These represent the values ​​before and after removal. The residual matrix of the thermal image sequence matrix and response matrix after the information of each latent variable has been explained; For the first The weight vector of the thermal image sequence matrix corresponding to each latent variable; For the first The response variable weight vector corresponding to the latent variable; and The input matrix and response matrix are respectively at the th The score vector in the direction of each latent variable.

[0060] No. The objective of extracting the latent variables can be expressed as:

[0061] Therefore, the first The latent variables represent the values ​​before elimination. Even after the influence of a latent variable, there is still a potential direction of change with a large covariance between the input thermal image sequence matrix and the response matrix. In other words, the "nth latent variable" represents the extraction order in partial least squares decomposition, rather than a fixed time point, a frame image, or a frequency component.

[0062] The matrix factorization form of partial least squares can be expressed as:

[0063]

[0064] in, Thermal image sequence matrix The latent variable score matrix; For response matrix The latent variable score matrix; The load matrix is ​​the input matrix; The load matrix is ​​the response matrix; and These are the residual matrices of the input matrix and the response matrix, respectively. This represents the number of latent variables extracted.

[0065] Score matrix and A linear relationship can be established between them:

[0066] in, This is the coefficient matrix in the regression relationship. Let be the residual matrix.

[0067] The spatial weight matrix corresponding to the input matrix is ​​denoted as:

[0068] in, No. Spatial weight vectors corresponding to each latent variable.

[0069] Due to the input matrix Each column corresponds to a fixed spatial pixel in the thermal image, therefore Each element corresponds to a pixel position in the original image.

[0070] Let the first The pixel at the th point The spatial weights of the latent variables are: Then the first The spatial weight vector corresponding to each latent variable is represented as follows: ,in, express The i-th pixel is at the i-th position The magnitude of the contribution in each latent variable direction. In other words, the spatial weight vector. This describes the temperature residual response of each pixel and the reference response matrix. The spatial distribution characteristics of the synergistic change relationship between them.

[0071] It should be noted that the spatial weights obtained by partial least squares are not the actual temperature value at a certain moment, nor are they phase information in the frequency domain. Instead, they are the spatial contribution coefficients of the input thermal image sequence in the direction of the latent variable. Since the hollow area and the normal area differ in thermal resistance conditions and heat transfer paths, their surface temperature response may exhibit different response amplitudes and rates of change during heating, cooling, and thermal state transitions. Therefore, the spatial weight feature map can, to some extent, highlight the spatial response differences related to local thermal resistance anomalies.

[0072] In summary, the input for this step is: and The output is the partial least squares decomposition result, especially the spatial weight matrix corresponding to the input matrix. .

[0073] 5. Extract the first latent variable space weight vector After partial least squares decomposition, the spatial weight vectors corresponding to each latent variable can be obtained:

[0074] The spatial weight vector corresponding to the first latent variable is:

[0075] The first latent variable is in the input matrix. With response matrix The main potential components are extracted from the directions where the synergistic changes are most significant. Therefore, the corresponding spatial weight vector can reflect the main spatial response patterns that are strongly related to the unsteady thermal processes of the wall within the detection time window.

[0076] For hollow defects, local thermal resistance anomalies alter the amplitude and rate of change of the temperature rise and fall response in the defect region. When the defect's thermal response is strongly coupled with the overall thermal excitation process, its anomalous characteristics are often clearly reflected in the first latent variable spatial weight vector. Compared to subsequent latent variables, the first latent variable usually contains more important and stable information about coordinated changes, and is relatively less affected by random noise and local interference. Therefore, it is more suitable as the primary basis for the initial identification and feature enhancement of hollow defects.

[0077] It is important to emphasize that the first latent variable is not necessarily identical to defect information, but rather prioritizes the spatial response pattern in the input thermal image sequence that changes most significantly in tandem with the thermal process reference response matrix. After dynamic background subtraction reduces the overall background temperature drift, this spatial response pattern is more effective in highlighting the differentiated thermal response caused by local thermal resistance anomalies.

[0078] In summary, the input for this step is the spatial weight matrix. The output is the spatial weight vector corresponding to the first latent variable. .

[0079] 6. Feature Image Generation Due to the weight vector of the first latent variable space The length is equal to the total number of pixels in a single frame of thermal image. Therefore, it can be restored to a two-dimensional matrix according to the spatial dimensions of the original thermal image, thus obtaining the spatial weight feature map of the first latent variable. This can be expressed as a formula:

[0080] in, Represents the first in the original image line, number The spatial weight value of the column pixel in the direction of the first latent variable.

[0081] If a column-first expansion method is used, the one-dimensional index of the pixel... with two-dimensional coordinates The correspondence is as follows:

[0082] At this point:

[0083] Since the sign direction of spatial weights in partial least squares is somewhat arbitrary, the positive or negative direction of the same latent variable does not affect the cooperative change relationship it represents. For the identification of unknown defect regions, to avoid relying on known defect locations to determine the positive or negative polarity of the image, a spatial weighted anomaly intensity map can be further constructed.

[0084] set up express If the median of all pixel values ​​is used, then the anomaly deviation strength of the spatial weights of the first latent variable can be expressed as:

[0085] in, express The deviation of each pixel from the overall spatial weight background is determined by this process, which does not depend on the known location of the defect area and is suitable for enhanced identification of unknown hollow defects.

[0086] Subsequently, Normalization is performed:

[0087] in, It is a very small positive number, used to avoid the denominator being zero; If an 8-bit grayscale image needs to be generated, it is further mapped as follows:

[0088] The final image obtained is the feature-enhanced image of the first latent variable:

[0089] The image reflects the intensity of abnormal response at each spatial location in the direction of the first latent variable. Areas with higher gray values ​​indicate that the location deviates significantly from the overall spatial weight background and can be used as suspected locations of hollow defects for subsequent interpretation or quantitative evaluation.

[0090] In summary, the input for this step is the weight vector of the first latent variable space. The output is the normalized first latent variable feature enhancement image. .

[0091] Example 2 One embodiment of the present invention provides a feature enhancement and construction system for external wall hollow defects based on time-lapse thermal imaging, comprising: The determination module is configured to: determine the optimal dynamic detection time window and sampling interval parameters based on the diurnal evolution of the wall orientation and surface temperature field of the target building's exterior wall; The acquisition module is configured to acquire a continuous three-dimensional thermal image sequence of the exterior wall of the target building, wherein the sequence is obtained by taking continuous images of the exterior wall of the target building at equal time intervals based on time-lapse thermal imaging and using a sampling interval duration parameter; The interception module is configured to: intercept a short-time continuous thermal response sequence of corresponding length from the continuous three-dimensional thermal image sequence according to the start and end instructions of the dynamic detection time window; The construction module is configured to: perform dimensionality reduction, transformation and feature enhancement processing on the extracted short-time continuous thermal response sequence based on partial least squares method to construct a feature image characterizing the hollow defects of the external wall.

[0092] Example 3 One embodiment of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for enhancing the features of external wall hollow defects based on time-delayed thermal imaging.

[0093] Example 4 In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, the method for enhancing the construction of external wall void defect features based on time-delayed thermal imaging is implemented.

[0094] Example 5 One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for enhancing the features of external wall voids based on time-delayed thermal imaging.

[0095] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for enhancing and constructing features of external wall hollow defects based on time-lapse thermal imaging, characterized in that, include: Based on the diurnal evolution of the wall orientation and surface temperature field of the target building's exterior wall, the optimal dynamic detection time window and sampling interval parameters are determined. A continuous three-dimensional thermal image sequence of the exterior wall of the target building is obtained. The sequence is based on time-lapse thermal imaging and is obtained by taking continuous images of the exterior wall of the target building at equal time intervals using a sampling interval parameter. Based on the start and end instructions of the dynamic detection time window, a short-time continuous thermal response sequence of corresponding length is extracted from the continuous three-dimensional thermal image sequence; Based on partial least squares, the short-time continuous thermal response sequence is subjected to dimensionality reduction, transformation and feature enhancement to construct a feature image characterizing the hollow defects of the exterior wall.

2. The method for enhancing and constructing features of external wall hollow defects based on time-lapse thermal imaging as described in claim 1, characterized in that, The dynamic detection time window adopts the extreme value neighborhood time window, which is defined as: based on the daily cycle evolution law of the surface temperature field, and according to real weather or test curves, tracking and locking the natural physical moment when the surface temperature of the target building's exterior wall reaches the highest or lowest temperature extreme value of the day, and constructing a local short time sequence with this extreme value moment as the center point and extending equidistantly before and after it.

3. The method for enhancing and constructing external wall hollow defect features based on time-delay thermal imaging as described in claim 2, characterized in that, The extreme value neighborhood time window is adjusted according to the wall orientation, with the total span duration of the window adjusted accordingly.

4. The method for enhancing and constructing external wall hollow defect features based on time-delayed thermal imaging as described in claim 3, characterized in that, The total span duration of the adjustment window is as follows: For the east wall, south wall, and west wall, the total span of the extreme value neighborhood time window is set to N hours around the neighborhood of the extreme temperature moment, where N is 3 hours or 4 hours. For the north-facing wall facing away from the sun, the total span of the extreme value neighborhood time window is set to a full-cycle time window that includes a complete 24-hour day and night.

5. The method for enhancing and constructing features of external wall hollow defects based on time-delayed thermal imaging as described in claim 1, characterized in that, The sampling interval duration parameter is used to implement zoned control of the shooting frequency: For the east wall, south wall, and west wall, the sampling interval duration parameter is set to M minutes, where M is any integer value less than 30 minutes; For the north wall, the sampling interval duration parameter is set to L minutes, where L is any integer value not greater than 10 minutes.

6. The method for enhancing and constructing external wall hollow defect features based on time-delayed thermal imaging as described in claim 1, characterized in that, The partial least squares method is based on the first and second derivatives of the overall average temperature curve of the exterior wall within the current detection time window, which are used to construct a feature image.

7. A feature enhancement and construction system for external wall hollow defects based on time-lapse thermal imaging, characterized in that, include: The determination module is configured to: determine the optimal dynamic detection time window and sampling interval parameters based on the diurnal evolution of the wall orientation and surface temperature field of the target building's exterior wall; The acquisition module is configured to acquire a continuous three-dimensional thermal image sequence of the exterior wall of the target building, wherein the sequence is obtained by taking continuous images of the exterior wall of the target building at equal time intervals based on time-lapse thermal imaging and using a sampling interval duration parameter; The interception module is configured to: intercept a short-time continuous thermal response sequence of corresponding length from the continuous three-dimensional thermal image sequence according to the start and end instructions of the dynamic detection time window; The construction module is configured to: perform dimensionality reduction, transformation and feature enhancement processing on the extracted short-time continuous thermal response sequence based on partial least squares method to construct a feature image characterizing the hollow defects of the external wall.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for enhancing the construction of external wall hollow defect features based on time-delayed thermal imaging as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the method for enhancing the construction of external wall hollow defect features based on time-delayed thermal imaging as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the method for enhancing the features of external wall voids based on time-delayed thermal imaging as described in any one of claims 1-6.

Images