A method for identifying hollowing of external wall based on multiple isotope infrared scanning

Abstract

The application discloses a kind of outer wall hollow identification methods based on multiple same-phase infrared scanning, comprising the following steps: respectively at larger, smaller temperature difference period, the infrared video stream of outer wall is collected, and the RGB video stream of large temperature difference period is synchronously obtained;Based on RGB video stream, construct the three-dimensional model of outer wall, extract the highest temperature infrared image of large temperature difference period and the lowest temperature infrared image of small temperature difference period;Two infrared video streams are aligned with three-dimensional model, and the thermal response difference of the same position is compared;After comparing difference data, by inputting infrared image selected according to different temperature difference conditions, it is identified and labeled hollow area by machine learning model.The present application can effectively suppress the detection error introduced by the thermal physical property difference of material in single scanning, uneven solar radiation intensity and environmental temperature fluctuation through the space-time correlation analysis of two same-phase infrared video streams and RGB video streams.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of external wall hollow detection, and particularly relates to an external wall hollow identification method based on multiple homopolar infrared scanning. BACKGROUND

[0002] The importance of building external wall detection cannot be ignored. As the first line of defense of the building, the stability and safety of the external wall are directly related to the overall safety of the building. The cracking, water seepage and falling of the external wall not only affect the aesthetics of the building, but also pose a threat to the safety of the life and property of the residents. Therefore, regular detection of the condition of the external wall, timely discovery and repair of problems, have important significance for prolonging the service life of the building. Therefore, developing new detection technology, building a more efficient and intelligent detection system, and realizing the automatic identification and evaluation of external wall diseases have become an important direction of current research.

[0003] For intelligent detection technology, infrared thermal imaging technology has become an indispensable tool, especially in the field of external wall hollow detection, which has shown significant application value. Through analyzing the heat transfer speed and infrared imaging distribution, the technology can effectively identify the defects inside the external wall. The reason why infrared thermal imaging technology is valued in building external wall detection is that it can provide non-contact rapid scanning and large-area detection of infrared radiation anomalies on the external wall finish, so as to quickly locate the quality disease area of the external wall finish.

[0004] In the field of building external wall hollow detection, although infrared imaging technology has been widely used, its inherent drawbacks cannot be ignored. First of all, the application of this technology depends largely on the results of single scanning, and this single data acquisition method cannot fully reflect the dynamic changes of the thermal distribution of the external wall, thereby limiting the depth and breadth of the detection. Secondly, infrared imaging detection is easily disturbed by many external factors such as the thermal conductivity of the external wall material, the structure orientation, the environmental temperature and the humidity, and these variables lead to the instability of the thermal imaging data, increasing the complexity and uncertainty of image analysis. Furthermore, since the infrared imaging technology is sensitive to the identification threshold of the hollow, slight environmental changes or operation errors may cause deviation of the identification results, which to some extent reduces the accuracy of the detection results.

[0005] The prior art mainly relies on single scanning to detect hollowing, but its detection method relying on single scanning has certain deficiencies. The single scanning method is disturbed by various external factors in actual operation: different thermal conductivity and thermal capacity characteristics of the outer wall material can cause differences in the absorption and release of heat during thermal imaging, thereby affecting the accuracy of hollowing identification; the orientation of the building can affect the absorption of solar radiation, thereby changing the temperature distribution of the outer wall surface, making it difficult to maintain consistency of single scanning results; changes in external temperature, especially day-night temperature difference and seasonal temperature fluctuations, can also significantly affect the scanning results of infrared thermal imaging.

[0006] Therefore, the thermal imaging images of single scanning often have large differences, which may lead to misjudgment or missed detection in the image interpretation process, thereby reducing the overall accuracy of hollowing identification. This limitation shows that the application of infrared thermal imaging technology in building outer wall hollowing detection still needs to be further optimized and improved to overcome the influence of external environmental factors on the detection results and improve the stability and reliability of the technology. SUMMARY

[0007] The current infrared thermal imaging detection technology detects the hollowing of the building outer wall in a single detection. Due to factors such as the material, orientation, and external temperature of the outer wall, the accuracy of the outer wall hollowing identification is reduced. To solve this problem, the present application provides a method for identifying outer wall hollowing based on multiple simultaneous infrared scanning, which can effectively suppress the detection errors introduced by the differences in material thermal properties, uneven solar radiation intensity, and environmental temperature fluctuations in single scanning.

[0008] The purpose of the present application is achieved by the following technical solutions:

[0009] The present application provides a method for identifying outer wall hollowing based on multiple simultaneous infrared scanning, comprising the following steps:

[0010] S1 selects two time periods, one of which has a temperature difference between the maximum temperature and the minimum temperature greater than or equal to 10℃, as a larger temperature difference time period; the other time period has a temperature difference between the maximum temperature and the minimum temperature less than 10℃, as a smaller temperature difference time period;

[0011] In the larger temperature difference time period, infrared video stream collection and RGB video stream shooting of the outer wall are performed; in the smaller temperature difference time period, infrared video stream collection of the outer wall is performed;

[0012] S2 uses the RGB video stream shot in the larger temperature difference time period to perform three-dimensional reconstruction, generating a digital three-dimensional model of the outer wall;

[0013] Infrared images of the highest temperature moment are extracted from infrared video streams within a large temperature difference period as infrared images with a large temperature difference; infrared images of the lowest temperature moment are extracted from infrared video streams within a small temperature difference period as infrared images with a small temperature difference.

[0014] S3 uses infrared video streams with small and large temperature differences to spatially align with the digital 3D model of the exterior wall. After alignment, the thermal response differences at the same spatial location are compared.

[0015] Based on the aforementioned thermal response differences, S4 employs a machine learning-based image recognition model or image processing algorithm to identify and label the location and distribution range of external wall hollow defects by inputting image data selected according to temperature difference conditions from the aligned video stream.

[0016] Preferably, step S3, which compares the thermal response differences at the same location between infrared images with small and large temperature differences, specifically involves setting an appropriate threshold and using a threshold algorithm to compare the thermal response differences at the same location between infrared images with small and large temperature differences.

[0017] Preferably, the machine learning-based image recognition model is a convolutional neural network, which uses multiple infrared image datasets with known hollow locations and ranges as training samples to train the image recognition model and learn the feature relationship between thermal response differences and hollow areas.

[0018] Preferably, in step S4, when using a machine learning-based image recognition model to label the location and extent of the hollow area, the specific process is as follows:

[0019] S41 preprocesses the infrared images after alignment and thermal response difference analysis to remove noise, enhance the images, and normalize them. At the same time, it uses the 3D reconstruction results of RGB images to correct geometric deviations, ensuring the uniformity of infrared images in scale and brightness.

[0020] S42 initially separates the thermal response anomaly region using an image segmentation algorithm and extracts multi-scale features, including local temperature gradient, texture, and edges.

[0021] S43 constructs a multi-layer convolutional neural network, whose input is the preprocessed infrared image, and whose output is the boundary and location prediction of the hollow area.

[0022] S44 inputs the aligned infrared image into the trained CNN model, which automatically performs forward inference and outputs the location and extent of the hollow area.

[0023] Preferably, in step S4, the multi-layer convolutional neural network utilizes the infrared image dataset with hollow label in the training process, expands the samples through data enhancement, adopts smooth L1 loss and cross entropy loss to construct a multi-task loss function, and optimizes the parameters by using back propagation and gradient descent algorithm, and adjusts the hyperparameters by cross-validation.

[0024] Preferably, in step S4, the multi-layer convolutional neural network utilizes the infrared image dataset with hollow label in the training process, expands the samples through data enhancement, adopts smooth L1 loss and cross entropy loss to construct a multi-task loss function, and optimizes the parameters by using back propagation and gradient descent algorithm, and adjusts the hyperparameters by cross-validation.

[0025] The application also provides a wall hollow identification device based on the wall hollow identification method based on multiple homopolar infrared scanning.

[0026] The application also provides an electronic device comprising a processor and a memory, wherein the memory is used to store non-transitory computer instructions, and when the non-transitory computer instructions are executed by the processor, the processor implements the wall hollow identification method based on multiple homopolar infrared scanning.

[0027] The application also provides a storage medium for storing non-transitory computer instructions, and when the non-transitory computer instructions are executed, the wall hollow identification method based on multiple homopolar infrared scanning is executed.

[0028] Compared with the prior art, the application has the following advantages and beneficial effects:

[0029] (1) The wall hollow identification method based on multiple homopolar infrared scanning can effectively suppress the detection errors caused by the material thermal property difference in single scanning, uneven solar radiation intensity and environmental temperature fluctuation.

[0030] (2) The wall hollow identification method based on multiple homopolar infrared scanning utilizes the RGB image to perform three-dimensional reconstruction, and then obtains digital information modeling, thereby providing data support for the accurate alignment of subsequent infrared images.

[0031] (3) The outer wall hollow identification method based on multiple parity infrared scanning of the present application strictly limits the large temperature difference shooting condition, the small temperature difference scanning condition, and the large temperature difference scanning condition. The temperature changes in different time periods can highlight the thermal anomalies of the hollow area. When the temperature difference is large, the wall temperature is low due to heat dissipation at night, and the thermal hysteresis effect of the hollow area may make its temperature relatively high, thereby forming a clear thermal contrast. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The flowchart of the outer wall hollow identification method based on multiple parity infrared scanning of the embodiment of the present application. DETAILED DESCRIPTION

[0033] The present application will be further described in detail below in conjunction with the embodiments, but the implementation of the present application is not limited thereto.

[0034] EMBODIMENT

[0035] The outer wall hollow identification method based on multiple parity infrared scanning of the present embodiment includes the following steps:

[0036] S1 selects two time periods, one of which has a temperature difference between the maximum temperature and the minimum temperature greater than or equal to 10℃, as a large temperature difference time period; the other has a temperature difference between the maximum temperature and the minimum temperature less than 10℃, as a small temperature difference time period;

[0037] In the large temperature difference time period, infrared video stream acquisition and RGB video stream shooting are performed on the outer wall; in the small temperature difference time period, infrared video stream acquisition is performed on the outer wall;

[0038] S2 uses the RGB video stream shot in the large temperature difference time period to perform three-dimensional reconstruction, generating a digital three-dimensional model of the outer wall;

[0039] The infrared image at the highest temperature in the infrared video stream in the large temperature difference time period is extracted as a large temperature difference infrared image, and the infrared image at the lowest temperature in the infrared video stream in the small temperature difference time period is extracted as a small temperature difference infrared image;

[0040] S3 uses the small temperature difference infrared video stream and the large temperature difference infrared video stream to respectively perform spatial alignment with the digital three-dimensional model of the outer wall. After alignment, the thermal response difference of the same spatial position is compared;

[0041] S4, based on the thermal response difference, uses a machine learning-based image recognition model or image processing algorithm to input the image data selected according to the temperature difference condition in the aligned video stream, identify and label the position and distribution range of the outer wall hollow defects.

[0042] In step S4, when the position and range of the hollowing are labeled by using the image recognition model based on machine learning, the specific process is as follows:

[0043] In step S41, the infrared image after alignment and thermal response difference analysis is preprocessed to remove noise, enhance the image and normalize, and the three-dimensional reconstruction result of the RGB image is used to correct the geometric deviation, so as to ensure the uniformity of the image data in scale and brightness.

[0044] In step S42, the thermal response abnormal area is preliminarily separated by using an image segmentation algorithm, and local temperature gradient, texture and edge and other multi-scale features are extracted to provide sufficient information for deep learning.

[0045] In step S43, a multi-layer convolutional neural network is constructed, the input of which is the preprocessed infrared image, and the output is the boundary and position prediction of the hollowing area.

[0046] In the training process of the multi-layer convolutional neural network, an infrared image dataset with accurate hollowing labeling is used, and the samples are expanded by data enhancement, a multi-task loss function is constructed by using smooth L1 loss and cross-entropy loss, and the parameters are optimized by using back propagation and gradient descent algorithm, and the hyperparameters are optimized by cross-validation, so as to ensure that the model has good generalization ability and robustness.

[0047] In the model verification stage of the multi-layer convolutional neural network, the accuracy, recall rate and F1 score are evaluated by using an independent test set, the false detection and missed detection are optimized, and the intermediate feature map is used for visual analysis to confirm the effective capture of the hollowing area by the model.

[0048] In step S44, the aligned infrared image is input into the trained CNN model, the model automatically performs forward reasoning and outputs the position and range of the hollowing, and the redundant detection is eliminated by post-processing (such as non-maximum suppression) to ensure the accuracy of the result. The labeling result is presented in the form of a bounding box or a segmentation map on the RGB image.

[0049] The wall hollowing recognition device for realizing the wall hollowing recognition method based on multiple homopolar infrared scanning includes an infrared scanner and a camera, and can be installed on a drone with the infrared scanner.

[0050] The embodiment also provides an electronic device including a processor and a memory; the memory is used to store non-transitory computer instructions; when the non-transitory computer instructions are executed by the processor, the processor implements the wall hollowing recognition method based on multiple homopolar infrared scanning.

[0051] The embodiment also provides a storage medium for storing non-transitory computer instructions, which are executed to perform the wall hollowing recognition method based on multiple homopolar infrared scanning.

[0052] The method of the embodiment effectively solves the problem of inaccurate detection caused by single scanning affected by environmental factors in the prior art, overcomes the image position alignment problem caused by occlusion and inaccurate external parameters, and the error problem of multi-time point image splicing, thereby ensuring the accuracy of the outer wall hollow identification, significantly improving the reliability and accuracy of the detection, meeting the demand of high-precision data, providing an efficient solution for outer wall hollow detection in complex scenes, and realizing more accurate data acquisition.

[0053] The application can be widely applied to the fields of building outer wall hollow detection, energy-saving reconstruction, building maintenance and maintenance, and has important application value, especially in the outer wall detection of high-rise buildings, historical buildings and special environments.

[0054] The above embodiment is a preferred embodiment of the application, but the embodiment of the application is not limited by the above embodiment, and any change, modification, substitution, combination, simplification made without departing from the spirit and principle of the application should be an equivalent replacement method, and all are included in the protection scope of the application.

Claims

1. A method for identifying external wall hollowing based on multiple isotope infrared scanning, characterized in that, The method comprises the following steps: S1 selecting two time periods, one of which has a temperature difference between the maximum temperature and the minimum temperature greater than or equal to 10℃, as a large temperature difference time period; and the other of which has a temperature difference between the maximum temperature and the minimum temperature less than 10℃, as a small temperature difference time period; In the large temperature difference time period, infrared video stream acquisition and RGB video stream shooting are performed on the external wall; In the small temperature difference time period, infrared video stream acquisition is performed on the external wall; S2 generating a digital three-dimensional model of the external wall by three-dimensional reconstruction using the RGB video stream shot in the large temperature difference time period; extracting an infrared image at the moment of the highest temperature from the infrared video stream in the large temperature difference time period as a large temperature difference infrared image, and extracting an infrared image at the moment of the lowest temperature from the infrared video stream in the small temperature difference time period as a small temperature difference infrared image; S3 performing spatial alignment of the small temperature difference infrared video stream and the large temperature difference infrared video stream with the digital three-dimensional model of the external wall respectively, and after the alignment, comparing the thermal response differences at the same spatial positions; S4 based on the thermal response differences, using a machine learning-based image recognition model or an image processing algorithm, inputting the image data selected according to the temperature difference condition in the aligned video stream to recognize and label the positions and distribution ranges of the external wall hollow defects.

2. The method for identifying the hollow of external wall based on multiple homopolar infrared scanning according to claim 1, characterized in that, In step S3, the thermal response differences at the same positions of the small temperature difference infrared image and the large temperature difference infrared image are compared, specifically: a suitable threshold is set, and a threshold algorithm is used to compare the thermal response differences at the same positions of the small temperature difference infrared image and the large temperature difference infrared image. 3.The external wall hollow identification method based on multiple homopolar infrared scanning according to claim 1, wherein, The machine learning-based image recognition model is a convolutional neural network, and a plurality of infrared image data sets with known hollow position and range annotations are used as training samples to train the image recognition model and learn the feature relationship between the thermal response differences and the hollows.

4. The method for identifying the hollowed outer wall based on multiple homopolar infrared scanning according to claim 1, characterized in that, In step S4, when the machine learning-based image recognition model is used to label the positions and ranges of the hollows, the specific process is as follows: S41 pre-processing the infrared image after alignment and thermal response difference analysis to remove noise, enhance the image and normalize, and correcting geometric deviation by using the RGB image three-dimensional reconstruction result to ensure the uniformity of the infrared image in scale and brightness; S42 preliminarily separating the thermal response abnormal area by an image segmentation algorithm and extracting multi-scale features, the multi-scale features including local temperature gradient, texture and edge; S43 constructing a multi-layer convolutional neural network, the input of which is the pre-processed infrared image, and the output is the boundary and position prediction of the hollow area; S44 inputting the aligned infrared image into the trained CNN model, and the model automatically performs forward inference and outputs the hollow position and range.

5. The method for identifying the external wall voids based on multiple homopolar infrared scanning according to claim 4, characterized in that, In step S4, the multi-layer convolutional neural network uses infrared image data sets with hollow annotations in the training process, and expands the samples by data augmentation, simultaneously constructs a multi-task loss function by using smooth L1 loss and cross-entropy loss, and then optimizes the parameters by using back propagation and gradient descent algorithm, and adjusts the hyperparameters by cross-validation.

6. The method for identifying the hollowed outer wall based on multiple homopolar infrared scanning according to claim 4, characterized in that, In step S4, the multi-layer convolutional neural network uses an independent test set to evaluate the accuracy, recall rate and F1 score in the model verification stage, and the false detection and missed detection are optimized, and the intermediate feature map is used for visual analysis to confirm the effective capture of the hollow area by the model.

7. An electronic device, comprising: comprising a processor and a memory: the memory is configured to store non-transitory computer instructions; when the non-transitory computer instructions are executed by the processor, the processor implements the method for identifying external wall hollowing based on multiple parity infrared scanning according to any one of claims 1-6.

8. A storage medium, characterized by a non-transitory computer instruction storage, when the non-transitory computer instructions are executed, the method for identifying external wall hollowing based on multiple parity infrared scanning according to any one of claims 1-6 is executed.

Images