External wall quality defect identification method and system

Abstract

The invention provides an external wall quality defect identification system, which comprises a preprocessing module used for preprocessing an image of a to-be-detected external wall, the image comprising an infrared image and a high-definition image of the external wall; the defect judgment module is used for analyzing the image by using a pre-trained machine learning model so as to identify and mark potential defects in the image; the defect confirmation module is used for finely screening the potential defects so as to determine real defects from the potential defects; and the output module is used for outputting the image of the outer wall marked with the defect. The invention further provides a corresponding method.

Description

Technical Field

[0001] The present invention relates to the technical field of building detection, and more specifically to a method and system for identifying exterior wall quality defects. Background Art

[0002] As the use time increases, various quality defects may appear in the building exterior walls due to factors such as environmental weather and construction quality. These defects may cause problems such as reduced insulation effect and wall falling off, which will seriously affect the service life and safety of the building.

[0003] The traditional exterior wall inspection method is that inspectors use handheld infrared thermal line inspection instruments to visually identify wall defects. Due to equipment limitations, this method is only suitable for inspecting low-rise exterior walls, and the judgment of defects is completely dependent on the experience of inspectors. It not only consumes a lot of manpower and time costs, but also has very limited detection precision and accuracy. In recent years, some exterior wall inspection methods have begun to use machine learning models to identify defects in exterior wall images to improve recognition accuracy and efficiency. However, these machine learning models are simply trained to identify specific types of defects, and their recognition accuracy is also determined by the matching degree between the training set and the defects in the inspected exterior wall, that is, only when detecting the corresponding type of defects can a good accuracy be achieved. In addition, these methods do not take into account the impact of actual factors such as ambient temperature and exterior wall structure on defect recognition, resulting in high recognition error rate and poor reliability.

[0004] Therefore, there is an urgent need for a method and system that can efficiently, accurately and reliably identify exterior wall quality defects. Summary of the invention

[0005] The present invention is intended to overcome the above and / or other problems of the prior art. The exterior wall quality defect identification method and system provided by the present invention utilizes both infrared images and high-definition images of the exterior wall to identify potential defects based on a machine learning model, and then screens out real defects from potential defects, thereby identifying exterior wall defects with high precision and efficiency while also taking into account the influence of actual factors such as ambient temperature and exterior wall structure, so that the identification results are quite accurate and reliable. In addition, the above method and system of the present invention can also be combined with manual verification to further improve the recognition accuracy.

[0006] According to a first aspect of the present invention, a system for identifying defects in exterior wall quality is provided, the system comprising a preprocessing module, a defect judgment module, a defect confirmation module and an output module. The preprocessing module is used to preprocess an image of an exterior wall to be inspected, the image comprising an infrared image and a high-definition image of the exterior wall. The defect judgment module uses a pre-trained machine learning model to analyze the image to identify and annotate potential defects in the image. The defect confirmation module is used to finely screen the potential defects to determine the real defects from the potential defects. The output module is used to output the image of the exterior wall annotated with the defects.

[0007] According to a second aspect of the present invention, a method for identifying exterior wall quality defects is provided, comprising the following steps: a) collecting images of the exterior wall to be inspected by a drone, the images including infrared images and high-definition images of the exterior wall; b) preprocessing the images; c) analyzing the images using a pre-trained machine learning model to identify and annotate potential defects in the images; d) finely screening the potential defects to determine real defects from the potential defects; and e) outputting the image of the exterior wall annotated with the defects.

[0008] Unlike the existing exterior wall defect recognition methods, the present invention can analyze both the infrared image and the high-definition image of the exterior wall at the same time, so that more types of defects can be recognized and the recognition accuracy can be significantly improved. Moreover, the present invention creatively uses the machine learning model to preliminarily identify the potential defects in the exterior wall image, and then finely screens out the real defects from the potential defects, thereby significantly improving the recognition accuracy and retaining the high precision and high efficiency unique to the machine learning model. The present invention can also directly mark the identified defects in the exterior wall image, so that the recognition results can be intuitively displayed, which is convenient for subsequent verification or use.

[0009] Optionally, the preprocessing module in the exterior wall quality defect recognition system of the present invention may be further configured to normalize the infrared image according to the ambient temperature at the time of shooting, and the defect judgment module may be further configured to take into account the ambient temperature at the time of shooting the infrared image during analysis, so as to take into account the influence of the ambient temperature when judging potential defects. Correspondingly, the exterior wall quality defect recognition method of the present invention may include normalizing the infrared image according to the ambient temperature at the time of shooting in the preprocessing step, and taking into account the ambient temperature at the time of shooting the infrared image during analysis.

[0010] Optionally, the defects detected by the exterior wall quality defect recognition system of the present invention may include hollowing, leakage, cracks, breakage and falling off, etc. These defects can be identified in high-definition images and / or infrared images.

[0011] Optionally, the fine screening may include one or more of the following: using the architectural drawings of the exterior wall to screen the potential defects, thereby eliminating misjudgments caused by factors such as building materials; using the corresponding high-definition images and infrared images collected at the same location to screen the potential defects, because the information in one type of image may be used to eliminate or confirm defects in another type of image; taking into account the impact of thermal radiation on the infrared image; and presenting the annotated images to professionals for manual judgment.

[0012] Optionally, the defect confirmation module in the exterior wall quality defect identification system of the present invention can feed back the determined real defects to the defect judgment module for further training of the machine learning model. Correspondingly, the exterior wall quality defect identification method of the present invention can further include the following steps: f) further training the machine learning model using the determined real defects.

[0013] Optionally, the exterior wall quality defect identification system of the present invention may further include a prediction module for predicting the development trend of the defect based on historical data. Correspondingly, the exterior wall quality defect identification method of the present invention may further include the following steps: g) predicting the development trend of the defect based on historical data.

[0014] According to a third aspect of the present invention, there is further provided a computer-readable storage medium having instructions stored thereon, which implement the above-mentioned exterior wall quality defect identification method when the instructions are executed.

[0015] Other features and aspects of the present invention will become more apparent from the following detailed description in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention may be better understood by describing exemplary embodiments of the present invention in conjunction with the accompanying drawings, in which:

[0017] Figure 1 A schematic block diagram of an exterior wall quality defect identification system according to the present invention is shown;

[0018] Figure 2 A flow chart showing a method for identifying exterior wall quality defects according to the present invention is shown;

[0019] Figure 3 A flow chart showing a variant embodiment of the method for identifying exterior wall quality defects according to the present invention is shown;

[0020] Figure 4 An example of an exterior wall image in which defects can be identified using the exterior wall quality defect identification system according to the present invention is shown;

[0021] Figure 5An example of an exterior wall image in which defects can be identified using the exterior wall quality defect identification system according to the present invention is shown;

[0022] Figure 6 An example of an exterior wall image in which defects can be identified using the exterior wall quality defect identification system according to the present invention is shown;

[0023] Figure 7 An example of an exterior wall image showing defects that can be identified using the exterior wall quality defect identification system according to the present invention; and

[0024] 8( a ), 8 ( b ), and 8 ( c ) respectively show examples of exterior wall images in which defects can be identified using the exterior wall quality defect identification system according to the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with specific embodiments and drawings. More details are elaborated in the following description to facilitate a full understanding of the present invention. However, the present invention can obviously be implemented in a variety of other ways different from the description herein. Those skilled in the art can make similar generalizations and deductions based on actual application situations without violating the spirit of the present invention. Therefore, the scope of protection of the present invention should not be limited by the content of this specific embodiment.

[0026] Unless otherwise defined, the technical or scientific terms used in the claims and the specification shall have the usual meaning understood by persons with ordinary skills in the technical field to which the invention belongs. The words "first", "second" and similar words used in the specification and claims of this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "one" or "a" do not indicate a quantitative limitation, but rather indicate the presence of at least one. Words such as "include" or "comprise" and similar words mean that the elements or objects appearing before "include" or "comprises" cover the elements or objects listed after "include" or "comprises" and their equivalent elements, and do not exclude other elements or objects.

[0027] According to an embodiment of the present invention, a system for identifying exterior wall quality defects is provided.

[0028] refer to Figure 1 , which shows an exterior wall quality defect identification system 100 according to the present invention, which includes a preprocessing module 110, a defect judgment module 120, a defect confirmation module 130 and an output module 140.

[0029] The preprocessing module 110 can be used to preprocess the image of the exterior wall to be inspected. Specifically, the preprocessing operation may include (but is not limited to) any one or more of the following operations: screening the image to eliminate images that are not suitable for inspection (e.g., images that are severely out of focus or too blurred, etc.); cropping and scaling the image to obtain an image size suitable for processing and / or removing the portion of the image that does not contain the exterior wall; transforming the image by image processing technology (e.g., rotating, flipping, translating, contrast and brightness adjustment, etc.) to enhance the information of interest in the image, etc. By performing preprocessing on the image, the exterior wall portion of interest in the image can be highlighted and the image size can be made more suitable for processing by subsequent modules, thereby improving the efficiency of the system 100 in identifying defects. Moreover, by eliminating images that are not suitable for inspection, ineffective workload can also be avoided.

[0030] The images of the exterior wall detectable by the system 100 may include infrared images and high-definition images, and optionally any other type of images that are helpful in identifying defects. For example, the images may be taken manually or by a drone. Preferably, a drone equipped with a camera (e.g., a thermal imaging camera and a high-definition camera, etc.) may be used to take both infrared images and high-definition images of the exterior wall (as well as optional other types of images) at (multiple) locations, thereby easily acquiring images of the exterior walls of high-rise buildings. In some embodiments, since there may be a small difference in the position / shooting angle of a corresponding set of infrared images and high-definition images of the exterior wall, the pre-processing module 110 may be further configured to transform the images so that the fields of view in the corresponding infrared images and high-definition images are more consistent.

[0031] The defect judgment module 120 can use a pre-trained machine learning model to analyze the exterior wall image to identify and mark potential defects in the image. In some embodiments, a machine learning model can be pre-trained for each type of image (such as infrared images, high-definition images, etc.), and then the defect judgment module 120 can select a corresponding machine learning model to analyze the corresponding type of image. Since the defects presented in the same type of exterior wall images usually have similar characteristics, a higher defect recognition accuracy can be obtained by training the machine learning model according to the image type. In some embodiments, according to actual conditions, it is also possible to choose to train the same machine learning model for multiple similar types of images, or to train multiple machine learning models for different defect types in the same type of images.

[0032] According to a specific embodiment of the present invention, the machine learning model used may be a model based on the YOLO (You Only Look Once) detection algorithm, which can achieve accurate segmentation and recognition of images. It should be understood that the machine learning model can also be constructed based on any suitable image recognition algorithm. By using the machine learning model for preliminary recognition, the defect judgment module 120 can accurately and quickly identify areas in the image where defects may exist, and mark these areas as potential defects for further confirmation.

[0033] The defect confirmation module 130 can be used to finely screen potential defects to determine real defects from potential defects. Due to the influence of various factors such as false positives of machine learning models, specific structures and materials in the exterior wall, and the ambient temperature when the picture is taken, there may be some misjudgments in the potential defects identified by the defect judgment module 120. The defect confirmation module 130 can eliminate misjudgments by performing one or more rounds of screening on potential defects, where each round of screening can be dedicated to eliminating the influence of different factors. After screening by the defect confirmation module 130, the potential defects that are finally retained will be defects with high confidence, that is, real defects.

[0034] The output module 140 may be used to output an image of an exterior wall with defects marked. For example, the output module 140 may output an image of an exterior wall with real defects marked on a display or to a file, or output the image of the exterior wall in any other suitable form. In some embodiments, the output module 140 may also output an image of an exterior wall with potential defects marked, or output an image of an exterior wall with both potential defects and real defects marked. As needed, in some embodiments, the marked exterior wall image may be further edited or processed (e.g., calculating the area of ​​each defective area, etc.).

[0035] Compared with the existing technology that can only detect specific types of defects, the above-mentioned system of the present invention can simultaneously analyze multiple types of exterior wall images such as infrared images and high-definition images, and combine these images to accurately and precisely identify various types of defects, which is not possible with the existing technology. In addition, the above-mentioned system of the present invention uses a machine learning model to preliminarily identify potential defects, and then finely screens out real defects from potential defects, thereby significantly reducing misjudgments while maintaining high recognition accuracy and high recognition speed, and greatly improving the accuracy of the recognition results. In addition, by marking the identified defects in the exterior wall image for output, not only can the recognition results be displayed intuitively, but it is also convenient for further processing and utilization.

[0036] Optionally, the preprocessing module 110 may normalize the infrared image according to the ambient temperature at the time of shooting. In some embodiments, the ambient temperature data may be recorded while shooting the infrared image (for example, the ambient temperature data may be saved in the infrared image). In some other embodiments, the preprocessing module 110 may extract the background temperature value from the background portion in the infrared image as the ambient temperature. As a non-limiting example, the preprocessing module 110 may perform normalization by subtracting the temperature value of each pixel in the infrared image from the ambient temperature of the infrared image, so that the temperature value of each pixel in the normalized infrared image obtained represents the difference between it and the ambient temperature. It should be understood that any other suitable normalization method may also be selected to perform the normalization process. Since some types of exterior wall defects can cause the temperature of the defective area to be significantly higher or lower than the ambient temperature and / or the surrounding exterior wall area (for example, when the exterior wall surface is subjected to solar heat radiation, the temperature change of the hollow area is greater than that of the normal exterior wall; when rainwater seeps from the exterior wall surface, the temperature of the leaking area will be lower than that of the intact exterior wall), by normalizing the infrared image based on the ambient temperature, it is easier to qualitatively and quantitatively identify such defects, and the normalized image is also more conducive to the training and use of the machine learning model. In some embodiments, the preprocessing module 110 can also store the ambient temperature in association with the normalized infrared image.

[0037] Optionally, the defect judgment module 120 may take into account the ambient temperature when the infrared image was taken during analysis, thereby incorporating the effects of the ambient temperature when judging potential defects. For example, the defect judgment module 120 may input the infrared image and the associated ambient temperature together into a machine learning model pre-trained based on the infrared image combined with the ambient temperature to identify potential defects. In order to utilize the obvious temperature difference between defective areas such as hollows and leaks and the surrounding normal exterior walls to judge such defects, a reference temperature difference may be set for each of such defects, and then the reference temperature difference (or a certain multiple of the reference temperature difference) is used as a threshold for judging whether such defects exist in the exterior wall area. For example, when the temperature difference exceeds the threshold range of a certain defect, it can be judged that the area has a potential defect of that type.

[0038] As a specific embodiment, one or more high-temperature points can be selected from the normalized infrared image as reference points. These reference points need to be verified (for example, manually verified using a hollowing hammer) to show that there are no hollowing defects. Then, the difference between the (average) temperature of the reference point and the ambient temperature is used as the reference temperature difference, and the threshold for judging hollowing defects is set according to the percentage of deviation from the reference temperature difference (for example, a hollowing defect is considered to exist if the temperature difference is greater than 30% of the reference temperature difference (i.e., greater than 1.3 times the reference temperature difference)); or alternatively, the threshold range can be set based on experience. However, the ambient temperature itself also has an impact on the temperature difference of the defective area. For example, for the same exterior wall, at different ambient temperatures, the defective area may show different temperature differences (for example, the size of the temperature difference may increase with the increase of the ambient temperature). At this time, if the same threshold is still used, it may cause drift in the judgment of the same potential defect at different ambient temperatures. According to a specific embodiment of the present invention, the defect judgment module 120 can adjust the threshold range accordingly according to the ambient temperature when the infrared image is taken (for example, enlarging the empirical threshold range when the ambient temperature is high, or increasing the deviation percentage of the reference temperature difference used when setting the threshold), thereby effectively reducing such misjudgments.

[0039] Defects that can be identified by the exterior wall quality defect identification system 100 may include: hollowing, leakage, cracks, breakage, falling off, and any other defects that can be identified in an image (e.g., a high-definition image, an infrared image, etc.). For example, hollowing and leakage areas may show a different temperature from the surrounding exterior wall areas in an infrared image; cracks, breakage and falling off areas may show obvious features in a high-definition image. In some cases, certain defects may be identified in different types of images. For example, hollowing and leakage defects can be identified in some infrared images and high-definition images at the same time. In some cases, certain defects may not be identified in a certain type of image. For example, a crack area may have the same temperature as the surrounding wall surface and therefore cannot be identified in an infrared image, but can be found in a high-definition image.

[0040] refer to Figures 4 to 7 , showing several examples of defects that can be identified using the exterior wall quality defect identification system according to the present invention. Figure 4 An infrared image (left) and a high-definition image (right) of an exterior wall with annotated hollowing defects are shown, which appear as hot spots (i.e., red areas indicating higher temperatures than the surrounding areas) in the infrared image and bulging of the wall in the high-definition image. Figure 5 An infrared image (left) and a high-definition image (right) of an exterior wall with annotated leakage defects are shown, which appear as cold spots (i.e., green areas indicating lower temperatures than surrounding areas) in the infrared image and wall damage in the high-definition image. Figure 6An infrared image (left) and a high-definition image (right) of an exterior wall with a leakage defect are shown. The defect appears as a large cold spot in the infrared image, but there is no obvious abnormality in the high-definition image. Figure 7 An infrared image (left) and a high-definition image (right) of an exterior wall with a parapet drain hole blockage defect are shown. The defect is manifested as a hot spot in the infrared image and a wall deformation in the high-definition image.

[0041] Optionally, the defect confirmation module 130 may use the architectural drawings of the exterior wall to screen potential defects. The defect confirmation module 130 may take into account the impact of the building structure and material on various images, for example, the infill wall, concrete wall and steel bar have a greater impact on temperature, the waterproofing coil has a relatively regular shape, the waterproofing coating has an irregular shape, etc. The defect confirmation module 130 may also exclude misjudgments caused by factors such as hollow parts and shadows by analyzing the building structure in the drawings.

[0042] refer to Figure 8(a) to Figure 8(c) , shows several examples of potential defects that can be screened using architectural drawings. Figure 8(a) to Figure 8(c) A set of infrared images (left) and high-definition images (right) of the exterior wall are shown in FIG.

[0043] For Figure 8(a), there are multiple irregular high-temperature hot spots (i.e., gray-white areas) and multiple rectangular hot spots in the infrared image of the exterior wall that may be identified as potential defects by the defect judgment module 120. The defect confirmation module 130 analyzes the architectural drawings and finds that the high-temperature hot spots correspond to the steel structure at the hollow part of the wall, which is at a significantly higher temperature due to its strong heat absorption capacity; the rectangular hot spots correspond to the insulation layer structure (similar rectangular hot spots also appear in Figures 8(b) and 8(c), which will not be repeated below), and therefore present a temperature different from the surrounding walls, so it can be determined that the potential defects are not real defects.

[0044] As for FIG8(b), there is a transversely elongated hot spot (corresponding to the black wall in the middle of the exterior wall in the high-definition image) in the infrared image of the exterior wall, which may be identified as a potential defect by the defect judgment module 120. The defect confirmation module 130 analyzes the architectural drawings and finds that the hot spot corresponds to a decorative waistline on the exterior surface of the wall, and therefore presents a temperature different from that of the surrounding walls, so it can be determined that the potential defect is not a real defect.

[0045] As for FIG8(c), there are three rows of long hot spots (corresponding to three rows of dark gray walls in the high-definition image) in the infrared image of the exterior wall, which may be identified as potential defects by the defect judgment module 120. The defect confirmation module 130 analyzes the architectural drawings and finds that the wall portion corresponding to the hot spots is made of lightweight block walls, and therefore presents a temperature different from that of the surrounding walls, so it can be determined that the potential defects are not real defects.

[0046] Optionally, the defect confirmation module 130 may analyze whether the exterior wall meets its design process based on the architectural drawings and the exterior wall image. Further, the defect confirmation module 130 may mark the exterior wall area that does not meet the design process as a defect.

[0047] Optionally, the defect confirmation module 130 may use the corresponding high-definition image and infrared image collected at the same location to screen potential defects. Additionally or alternatively, the defect confirmation module 130 may use other types of images collected at the same location to screen potential defects. For example, for an area identified as a potential defect in a certain type of image, the defect confirmation module 130 may obtain more information about the area by analyzing and comparing the area in other types of images corresponding thereto, and determine whether the potential defect is a real defect based on the information.

[0048] As a specific example, refer to Figure 4 , the defect judgment module 120 marks multiple hot spots in the infrared image as potential defects (marked with black boxes), and marks multiple uneven wall surfaces in the corresponding high-definition image as potential defects (marked with red boxes) (for clarity, Figure 4 Other defects are not marked in the image). The defect confirmation module 130 can compare the infrared image and the high-definition image marked by the defect judgment module 120, and find that the same potential defect area is identified in both types of images, and it has the characteristics of abnormally high temperature and uneven wall skin. Based on this, the defect confirmation module 130 can determine that these potential defects are real defects, and further, the defect confirmation module 130 can determine these defects as hollow defects based on the above characteristics. Similarly, when the information in a certain image can be used to determine that the potential defects in the corresponding other type of image are not real defects (for example, the high-definition image can indicate that the hot spots / cold spots in the corresponding infrared image are caused by metal components / shadows, such as the cold spots caused by the downward projection of the exterior wall decoration waistline in Figure 8(b), Figure 7 8(c) ), the defect confirmation module 130 may determine that the potential defect is a misjudgment.

[0049] Optionally, the defect confirmation module 130 may take into account the effect of thermal radiation on the infrared image. Since the exterior wall surface is affected by thermal radiation, the hot spots and cold spots in the infrared image may be larger or smaller than the actual defect area. The defect confirmation module 130 may correct the size of the defect area affected by thermal radiation (for example, based on the ambient temperature, etc.), so that the defect finally marked can correctly reflect the actual size of the defect area.

[0050] Optionally, the defect confirmation module 130 can also present the annotated image to professionals for manual judgment. For example, the defect confirmation module 130 can present the image annotated by the defect judgment module 120 on a display and allow professionals to confirm the potential defects in the image (for example, mark it as a real defect or delete it as a false positive, etc.). Since the defect judgment module 120 has already identified and annotated the potential defects in the image, professionals only need to judge the annotated potential defects without searching for defects in the entire image, which can significantly save workload compared to manual recognition of traditional methods. In addition, the introduction of manual judgment can also significantly improve recognition accuracy.

[0051] Optionally, the defect confirmation module 130 may feed back the determined real defects to the defect judgment module 120. In some embodiments, the defect judgment module 120 may update the training data set of the machine learning model based on the feedback data to retrain the machine learning model. In some embodiments, the defect judgment module 120 may periodically update the parameters of the machine learning model based on the feedback data.

[0052] Optionally, the exterior wall quality defect identification system 100 may also include a prediction module. When there are multiple sets of images and defect identification results collected across a certain length of time for the same exterior wall (for example, obtained after two or more inspections), the prediction module can predict the future development trend of the defects in the exterior wall based on these data. For example, the prediction module can model the changes in the length and width of crack defects over time to predict the future development trend of the crack. In some embodiments, the prediction module can also perform a temporal horizontal comparison of the same exterior wall to analyze the changes in defects in the same exterior wall area, such as the expansion, addition, or morphological changes of defects in the exterior wall area (for example, hollowing defects may cause cracking of the exterior wall over time, and the prediction module can judge that the hollowing defect is transformed into a crack defect). In a further embodiment, the prediction module can perform the above-mentioned horizontal comparison on the results of two or more optional inspections of the same exterior wall area.

[0053] According to an embodiment of the present invention, a method for identifying exterior wall quality defects is also provided accordingly.

[0054] refer to Figure 2 , which shows an exterior wall quality defect identification method 200 according to the present invention. The method 200 includes steps 210 to 250.

[0055] like Figure 2 As shown, in step 210, an image of the exterior wall to be inspected is collected by a drone, and the image may include an infrared image and a high-definition image of the exterior wall.

[0056] In step 220, the image is preprocessed. According to an embodiment of the present invention, the preprocessing may include but is not limited to filtering, cropping, scaling and transforming (eg, rotating, flipping, translating, adjusting contrast and brightness, etc.) the image.

[0057] In step 230 , the image is analyzed using a pre-trained machine learning model to identify and annotate potential defects in the image.

[0058] In step 240, the potential defects are finely screened to determine the real defects from the potential defects.

[0059] In step 250 , an image of the exterior wall with defects marked is output.

[0060] Optionally, the identified defects may include hollows, leaks, cracks, breaks, falls, and any other defects that can be identified in the image.

[0061] Optionally, the fine screening in step 240 may include one or more of the following: using architectural drawings of the exterior wall to screen potential defects, thereby eliminating misjudgments caused by factors such as building materials; using corresponding high-definition images and infrared images collected at the same location to screen potential defects, because specific information in one type of image may be used to eliminate some misjudgments in another type of image; taking into account the impact of thermal radiation on infrared images; and presenting the annotated images to professionals for manual judgment.

[0062] Optionally, step 220 may include normalizing the infrared image according to the ambient temperature at the time of taking the image, and taking into account the ambient temperature at the time of taking the infrared image during analysis.

[0063] refer to Figure 3 Optionally, the above-mentioned exterior wall quality defect identification method 200 may further include step 260 after step 250: further training the machine learning model using the determined real defects.

[0064] Optionally, the above-mentioned exterior wall quality defect identification method 200 may further include step 270 after step 250: predicting the development trend of the defects based on historical data.

[0065] The above-mentioned exterior wall quality defect identification method 200 corresponds to the exterior wall quality defect identification system according to the present invention. Many design concepts and details applicable to the above-mentioned exterior wall quality defect identification system of the present invention are also applicable to the above-mentioned exterior wall quality defect identification method 200, and the same beneficial technical effects can be obtained, which will not be repeated here.

[0066] According to an embodiment of the present invention, there is also provided a computer-readable storage medium having encoded instructions recorded thereon, which can implement the above-mentioned exterior wall quality defect identification method when the instructions are executed. The computer-readable storage medium may include a hard disk drive, a floppy disk drive, a compact disk read / write (CD-R / W) drive, a digital versatile disk (DVD) drive, a flash drive and / or a solid-state storage device, etc.

[0067] Various aspects of the present invention have been described above through some exemplary embodiments. However, it should be understood that various modifications may be made to the above exemplary embodiments without departing from the spirit and scope of the present invention. For example, if the described techniques are performed in a different order and / or if the components in the described system, architecture, device or circuit are combined in different ways and / or replaced or supplemented by other components or their equivalents, suitable results may be achieved, and accordingly, these modified other embodiments also fall within the scope of protection of the claims.

Claims

1. An exterior wall quality defect recognition system, comprising: A preprocessing module, used for preprocessing the image of the exterior wall to be detected, wherein the image includes an infrared image and a high-definition image of the exterior wall; a defect determination module, wherein the defect determination module uses a pre-trained machine learning model to analyze the image to identify and annotate potential defects in the image; A defect confirmation module, used for finely screening the potential defects to determine the real defects from the potential defects; as well as An output module is used to output the image of the exterior wall marked with the defects.

2. The system according to claim 1, characterized in that The pre-processing module is further configured to normalize the infrared image according to the ambient temperature when the image is taken, and the defect determination module is further configured to take into account the ambient temperature when the infrared image is taken during analysis.

3. The system according to claim 1, characterized in that in, The defects include hollowing, leakage, cracks, breakage and falling off.

4. The system according to claim 1, characterized in that The fine screening performed by the defect confirmation module includes one or more of the following: using the architectural drawings of the exterior wall to screen the potential defects; using the corresponding high-definition images and infrared images collected at the same location to screen the potential defects; taking into account the influence of thermal radiation on the infrared image; and presenting the annotated images to professionals for manual judgment.

5. The system according to claim 1, wherein: in, The defect confirmation module feeds back the determined real defects to the defect judgment module for further training of the machine learning model.

6. The system according to claim 1, wherein: The system further comprises a prediction module, which is used to predict the development trend of the defect according to historical data.

7. A method for identifying exterior wall quality defects, comprising the following steps: a) collecting images of the exterior wall to be inspected by a drone, wherein the images include infrared images and high-definition images of the exterior wall; b) preprocessing the image; c) analyzing the image using a pre-trained machine learning model to identify and annotate potential defects in the image; d) performing fine screening of the potential defects to determine the real defects from the potential defects; as well as e) outputting the image of the exterior wall marked with the defects.

8. The method according to claim 7, characterized in that The step b) further comprises: normalizing the infrared image according to the ambient temperature when the image was taken, and the step c) further comprises: taking into account the ambient temperature when the infrared image was taken during analysis.

9. The method according to claim 7, characterized in that in, The defects include hollowing, leakage, cracks, breakage and falling off.

10. The method according to claim 7, characterized in that The fine screening in step d) includes one or more of the following: using the architectural drawings of the exterior wall to screen the potential defects; using the corresponding high-definition images and infrared images collected at the same location to screen the potential defects; taking into account the influence of thermal radiation on the infrared image; and presenting the annotated images to professionals for manual judgment.

11. The method according to claim 7, characterized in that The method further comprises the following steps: f) Further training the machine learning model using the determined real defects.

12. The method according to claim 7, characterized in that The method further comprises the following steps: g) Predict the development trend of the defects based on historical data.

13. A computer-readable storage medium having instructions stored thereon, which implement the method according to any one of claims 7 to 12 when the instructions are executed.

Images