Machine learning-based external wall hollowing detection system and method

Abstract

The application provides a kind of machine learning-based outer wall hollow detection system and method, comprising: current outer wall image and environmental data of the day are collected, and matching image conforming to the environmental data of the day is found from pre-constructed outer wall image library;Outer wall image library stores the hollow image under complex environment and normal environment;Environmental data includes weather and temperature;The feature of current outer wall image and matching image conforming to the environmental data of the day is compared, and the initial judgment result is generated by analyzing whether the current outer wall is hollow combined with pre-constructed hollow judgment rule;Pre-constructed hollow judgment rule is obtained by analyzing the feature mode and feature change rule of outer wall hollow image based on pre-constructed outer wall image library;The knocking sound frequency of current outer wall is collected, to verify whether the initial judgment result is accurate, and generate outer wall detection result.The application can accurately identify outer wall hollow under various environmental conditions, greatly improve the reliability and efficiency of detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wall detection, in particular to a kind of outer wall hollowing detection method based on machine learning, a kind of outer wall hollowing detection system based on machine learning. BACKGROUND

[0002] Outer wall hollowing refers to the phenomenon that the decorative layer or base layer of the outer wall of the building is partially separated from the main structure to form a cavity; Hollowing can cause unevenness, cracks and other phenomena on the surface of the outer wall, seriously affecting the appearance of the building. Hollowing causes the formation of a cavity between the decorative layer and the base layer of the outer wall, which can easily accumulate water, accelerate the aging and corrosion of the wall material, and reduce the service life of the outer wall. Therefore, timely detection of outer wall hollowing is essential.

[0003] Existing detection methods include manual knocking, in which the detection personnel lightly tap the surface of the outer wall with a hammer or other tool and determine whether there is hollowing by sound and feel. If the sound is clear, it indicates that the wall surface is dense; If the sound is dull or there is a hollow sound, it may indicate that there is hollowing. This method is simple and easy to implement, but it is highly subjective and depends on the experience of the detection personnel, and it is inefficient and not suitable for large-area detection. More and more people have begun to use temperature distribution differences on the surface of objects to detect hollowing. The hollowing part will have a slight difference in temperature from the surrounding normal part due to air insulation. An infrared thermal imager can capture this difference and form a thermal image, based on which the location and range of hollowing can be analyzed quickly and conveniently. However, it is easily affected by weather conditions such as solar radiation, overcast weather, and heavy fog, which can reduce the temperature difference on the wall surface and result in a decrease in detection accuracy or even an inability to accurately detect. SUMMARY

[0004] In view of the above problems, the present application is proposed to provide a kind of outer wall hollowing detection method based on machine learning and corresponding outer wall hollowing detection system based on machine learning to overcome the above problems or at least partially solve the above problems.

[0005] The present application discloses an outer wall hollowing detection method based on machine learning, which comprises:

[0006] Collecting the current outer wall image and the environmental data of the day, and searching for a matching image that meets the environmental data of the day from a pre-constructed outer wall image library. The pre-constructed outer wall image library stores hollowing images under complex environments and hollowing images under normal environments. The environmental data includes weather and temperature.

[0007] Comparing the features of the current outer wall image and the matching image that meets the environmental data of the day, and analyzing whether the current outer wall has hollowing by combining a pre-constructed hollowing judgment rule to generate an initial judgment result. The pre-constructed hollowing judgment rule is constructed by analyzing the feature mode and feature change rule of the outer wall hollowing image based on the pre-constructed outer wall image library.

[0008] Collect the knocking sound frequency of the current outer wall, verify whether the initial judgment result is accurate, and generate an outer wall detection result.

[0009] Optionally, an outer wall image library is constructed, including:

[0010] The outer wall of the known hollow area is numbered and partitioned;

[0011] The complex environment data, the hollow image and the normal image under the complex environment, and the normal environment data, the hollow image and the normal image under the normal environment are collected for the hollow area and the normal area of the outer wall;

[0012] The collected images are sorted and classified, and each hollow image and normal image is associated with corresponding weather and temperature data.

[0013] Optionally, the characteristics mode and the characteristic change rule of the hollow image of the outer wall are analyzed based on the outer wall image library, and a hollow judgment rule is constructed, including:

[0014] The texture features of the complex environment hollow image and the normal environment hollow image are extracted respectively;

[0015] The difference between the texture features of the complex environment hollow image and the texture features of the normal environment hollow image is calculated to form a feature difference matrix, and the feature difference matrix is converted into a heat map;

[0016] The hollow images are grouped according to the weather and temperature, and the hollow image characteristic mode is calculated; the characteristic mode is the statistical value of the texture features of the hollow image;

[0017] The characteristic change rule is analyzed according to the heat map and the hollow image characteristic mode, and the hollow judgment rule is constructed based on the hollow image characteristic mode and the characteristic change rule of the outer wall.

[0018] Optionally,

[0019] The characteristic change rule is: compared with the normal environment, the dry and hot environment increases the feature difference between the hollow area and the normal area; the wet and cold environment reduces the feature difference between the hollow area and the normal area; the normal environment includes sunny and regular temperature; the dry and hot environment includes sunny and high temperature; the wet and cold environment includes rainy day, overcast day and low temperature;

[0020] The hollow judgment rule is: if the feature of the current outer wall image is small compared with the feature of the matching image according to the environmental data of the day, and at the same time meets the characteristic mode condition constraint of the hollow image in the characteristic change rule, it is determined that the current outer wall has hollow.

[0021] Optionally, the current outer wall image and the matching image conforming to the environmental data of the day are compared in feature, and combined with the pre-constructed hollow judgment rule, whether the current outer wall has hollow is analyzed to generate an initial judgment result, including:

[0022] The texture features of the current outer wall image and the matching image conforming to the environmental data of the day are extracted respectively, and the texture features of the normal image conforming to the environmental data of the day are extracted;

[0023] Based on the extracted texture features, the feature mode of the matching image conforming to the environmental data of the day is calculated, and the feature mode of the normal image conforming to the environmental data of the day is calculated;

[0024] The first difference value of the texture features of the current outer wall image and the feature mode of the matching image is calculated, and the second difference value of the texture features of the current outer wall image and the feature mode of the normal image is calculated;

[0025] If the first difference value is less than the second difference value, and the first difference value is less than the hollow difference value threshold, and the texture features of the current outer wall image meet the condition constraint based on the feature mode of the hollow image in the pre-constructed outer wall image library, an initial judgment result that the current outer wall has hollow is generated.

[0026] Optionally, the current outer wall image and the environmental data of the day are collected, and the matching image conforming to the environmental data of the day is searched from the pre-constructed outer wall image library, including:

[0027] The current outer wall image and the weather and temperature of the day are collected;

[0028] The weather and temperature of the day are taken as retrieval conditions to screen whether there is the same weather and temperature as the weather and temperature of the day in the pre-constructed outer wall image library;

[0029] If there is the same weather and temperature as the weather and temperature of the day, the image associated with the same weather and temperature is taken as the matching image conforming to the environmental data of the day;

[0030] If there is no weather and temperature completely same as the weather and temperature of the day, the image associated with the closest weather and temperature is selected as the matching image conforming to the environmental data of the day.

[0031] Optionally, the texture features include energy, entropy and contrast.

[0032] Optionally, the method further includes:

[0033] The gray level co-occurrence matrix of each image is established for the current outer wall image, the matching image conforming to the environmental data of the day and the normal image conforming to the environmental data of the day;

[0034] Energy, entropy, and contrast are calculated through the gray level co-occurrence matrix of each image.

[0035] Optionally, the knocking sound frequency of the current outer wall is collected to verify whether the initial judgment result is accurate, and an outer wall detection result is generated, including:

[0036] The knocking sound signal at the hollow position of the outer wall marked by the initial judgment result is collected, the knocking sound signal is preprocessed, and the frequency feature is extracted through frequency spectrum analysis;

[0037] The similarity between the frequency feature and the hollow sample in the hollow sound frequency database is calculated;

[0038] If the initial judgment result is that the outer wall is hollow, and the similarity is higher than the preset similarity threshold, an outer wall detection result that the current outer wall is hollow is generated.

[0039] The application also discloses an outer wall hollow detection system based on machine learning, which comprises:

[0040] An environmental data image matching module is configured to collect a current outer wall image and daily environmental data, and search for a matching image meeting the daily environmental data from a pre-constructed outer wall image library; the pre-constructed outer wall image library stores hollow images under complex environments and hollow images under normal environments; the environmental data includes weather and temperature;

[0041] An outer wall hollow initial judgment module is configured to compare features of the current outer wall image and the matching image meeting the daily environmental data, analyze whether the current outer wall is hollow in combination with a pre-constructed hollow judgment rule, and generate an initial judgment result; the pre-constructed hollow judgment rule is constructed by analyzing feature modes and change rules of the outer wall hollow images based on the pre-constructed outer wall image library;

[0042] An outer wall hollow verification module is configured to collect a knocking sound frequency of the current outer wall, verify whether the initial judgment result is accurate, and generate an outer wall detection result.

[0043] The application has the following advantages:

[0044] The outer wall hollow detection method based on machine learning can accurately identify the outer wall hollow under various environmental conditions, and greatly improves the reliability and efficiency of detection. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a step flowchart of the outer wall hollow detection method based on machine learning provided by the embodiment of the application;

[0046] Figure 2 is a structural block diagram of the outer wall hollow detection system based on machine learning provided by the embodiment of the application. DETAILED DESCRIPTION

[0047] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0048] Reference Figure 1 , a step flow chart of a machine learning-based external wall hollow detection method provided in an embodiment of the present application is shown, which can specifically include the following steps:

[0049] Step 101, current external wall images and daily environmental data are collected, and matching images conforming to the daily environmental data are searched from a pre-constructed external wall image library; the pre-constructed external wall image library stores hollow images under complex environments, hollow images under normal environments; the environmental data includes weather and temperature;

[0050] Step 102, feature comparison is performed on the current external wall images and the matching images conforming to the daily environmental data, and an initial judgment result is generated by analyzing whether the current external wall has hollows in combination with a pre-constructed hollow judgment rule; the pre-constructed hollow judgment rule is constructed by analyzing external wall hollow image feature patterns and feature change rules based on the pre-constructed external wall image library;

[0051] Step 103, the knocking sound frequency of the current external wall is collected to verify whether the initial judgment result is accurate, and an external wall detection result is generated.

[0052] The present application constructs an external wall image library by collecting infrared image data A of external wall hollows under complex weather and temperature conditions, infrared image data B of external wall hollows under normal weather and temperature, and external wall normal images under different weather and temperature conditions, which can comprehensively and meticulously collect external wall infrared image information under various conditions, provide rich and accurate data basis for subsequent analysis, and help to more accurately grasp the relationship between external wall hollows and weather and temperature. Moreover, the machine learning method is used to compare and analyze data A and data B, which can automatically explore the rules therein, and construct a hollow judgment rule based on the explored rules. Specifically, based on the rules explored in the comparison and analysis of data A and data B, the feature differences between newly collected images and similar images are judged, and the hollow judgment rule is: if the features of the current external wall images and the features of the matching images conforming to the daily environmental data are small, and the feature mode conditions of the hollow images in the feature change rule are met, it is preliminarily judged that the current wall has hollows; if the features are close to the feature mode of normal images, it is preliminarily judged that the current wall has no hollows. This intelligent analysis method not only improves the efficiency of data analysis, but also discovers some potential rules that are difficult for humans to detect, which provides a scientific basis for subsequent detection work.

[0053] In a specific detection process, the outer wall is detected by an infrared thermal imager, the infrared image of the outer wall can be quickly obtained, and the image corresponding to the temperature and weather of the day is quickly found in the constructed image library, and then the hollow judgment rule constructed according to the previously found rule is analyzed and judged to quickly obtain the preliminary result of whether the current wall body has a hollow condition, greatly improving the efficiency and accuracy of detection, and the outer wall hollow problem can be found in time, and timely measures can be taken for repair; and further, the unmanned aerial vehicle is used to carry a knocking hammer to knock the position of the preliminary result, and the sound frequency is obtained to verify the accuracy of the preliminary judgment result, the detection result is verified twice in this way, the reliability of the detection result is further improved, the possibility of misjudgment and omission is reduced, and the judgment of the outer wall hollow condition is ensured to be accurate.

[0054] In an optional embodiment of the present application, the outer wall image library is constructed, comprising:

[0055] The outer wall of the known hollow area is numbered and partitioned;

[0056] The hollow area and the normal area of the outer wall are collected, the complex environment data, the hollow image and the normal image under the complex environment, and the normal environment data, the hollow image and the normal image under the normal environment are collected;

[0057] The collected images are sorted and classified, and each hollow image and normal image is associated with corresponding weather and temperature data.

[0058] The infrared thermal imager and the environmental parameter measuring instrument are prepared in advance to provide professional tool support for data collection, the infrared thermal imager can accurately capture the temperature difference of the outer wall hollow, the environmental parameter measuring instrument can accurately record the temperature, humidity and other environmental factors, and the measurement error caused by equipment loss or temporary preparation is avoided, and the accuracy and reliability of the collected data are guaranteed from the source. The outer wall of the building with representative and known hollow area is numbered and partitioned, so that the collected data have a clear spatial correspondence, in the subsequent data sorting and analysis process, the data of the specific area can be quickly located, the data change of different areas and different times is compared, the efficiency of data management and analysis is greatly improved, and the relationship between the outer wall hollow and the environmental factors is more accurately explored.

[0059] The temperature, humidity and wind speed environment parameters and weather conditions of the day are recorded by using a thermometer and hygrometer and an anemometer. Data A is collected under different weather and temperature, covering sunny, cloudy, rainy, high temperature, low temperature and complex environment, and data B is collected under sunny weather and normal temperature, as a control data. Specifically, data collection is performed, and infrared and visible light photos are taken under the same distance and angle of the infrared thermal imager and the outer wall, and the environment parameters, time and position are recorded. The same collection steps as in sunny weather are adopted in cloudy weather. The collection and recording are performed in summer high temperature and winter low temperature periods. The image data collected is sorted and classified, and each hollow drum image and normal image is associated with the corresponding weather and temperature data.

[0060] In an optional embodiment of the present application, the characteristics and variation rules of the outer wall hollow drum image are analyzed based on the outer wall image library to construct a hollow drum judgment rule, which includes:

[0061] The texture features of the complex environment hollow drum image and the normal environment hollow drum image are extracted respectively;

[0062] The difference between the texture features of the complex environment hollow drum image and the normal environment hollow drum image is calculated to form a feature difference matrix, and the feature difference matrix is converted into a heat map;

[0063] The hollow drum images are grouped according to the weather and temperature, and the feature mode of the hollow drum image is calculated; the feature mode is the statistical value of the texture features of the hollow drum image;

[0064] The feature variation rules are analyzed according to the heat map and the feature mode of the hollow drum image, and the hollow drum judgment rule is constructed based on the feature mode and the feature variation rules of the outer wall hollow drum image.

[0065] The data A and B are preprocessed to remove interference and unify the standard, thereby ensuring the accuracy of the analysis. The texture features of the data A and B are extracted based on the image processing algorithm to mine the essential information of the hollow drum and improve the data value. The neural network model is constructed, the data A and B after processing are divided into a training set and a test set according to a proportion, the training set is used to train the model, the test set is used to evaluate the generalization ability of the model, and the model is optimized. The features extracted from the data A and B are compared, the difference between the features of the outer wall hollow drum infrared image and the normal environment under complex weather and temperature is analyzed, the distribution and variation rules of the features are displayed based on the heat map, the feature variation mode of the outer wall hollow drum area under different weather and temperature conditions is analyzed, the influence mechanism is mastered, and the system adaptability is enhanced; the data is integrated into the image library and labeled and classified, the database is improved, and the retrieval matching and system optimization are facilitated.

[0066] The present application firstly extracts texture features through a scientific and rigorous process when analyzing the texture features of target data, and standardizes the extracted texture features, which aims to eliminate the influence of the dimension and numerical range between different feature dimensions, so that each feature is in the same scale, facilitating subsequent analysis and comparison. After standardization, the difference between the features is calculated using a specific algorithm, and a feature difference matrix is constructed. The matrix records the difference between each texture feature in a quantitative form, providing an important data basis for subsequent research.

[0067] After obtaining the feature difference matrix, it is converted into an intuitive heat map with the help of professional visualization tools. During the drawing process, the horizontal and vertical axis information is carefully labeled, and the meaning of each axis is clearly defined. At the same time, a clear and eye-catching title is added, so that the viewer can quickly understand the content displayed by the heat map. The heat map visually displays the distribution of feature differences through changes in color depth. The darker the color, the greater the feature difference, and vice versa. This visualization method greatly improves the readability and understandability of the data.

[0068] In order to further explore the influence of different environmental factors on texture features, data A is grouped according to weather (sunny, cloudy) and temperature (high temperature, low temperature). For each group of hollow areas, a series of key feature statistics such as mean, variance, and standard deviation are calculated. These statistics reflect the central tendency and dispersion of the texture features of the hollow areas from different angles. Combined with the feature difference distribution shown in the heat map, the feature change trend under different groups is analyzed in detail to uncover the influence mechanism hidden in the data.

[0069] After systematic analysis, the following conclusions are obtained: compared with normal environment, dry and hot environment increases the feature difference between hollow area and normal area; wet and cold environment reduces the feature difference between hollow area and normal area; normal environment can include sunny and normal temperature; dry and hot environment can include sunny and high temperature; the wet and cold environment can include rainy, cloudy and low. Among them, the weather and temperature of the normal environment can generally refer to sunny and normal temperature, and the normal temperature is determined according to the actual time and space. Specifically, under the condition of high temperature and sunny weather, the texture feature difference is particularly significant; while in low temperature and cloudy weather, the feature difference is relatively small. Further research shows that strong light and high temperature environment in sunny weather can exacerbate the heat conduction difference between the hollow area and the normal area. Due to the change of heat conduction, the texture features of the object surface change significantly, which is fully reflected in the feature difference matrix and heat map. This research result has important significance for understanding the relationship between environmental factors and texture features, and can provide a powerful reference for related research and application.

[0070] In an optional embodiment of the present application, the current outer wall image and the matching image conforming to the environmental data of the day are compared in features, and the initial judgment result is generated by combining the pre-constructed hollow judgment rule to analyze whether the current outer wall has hollow, including:

[0071] The texture features of the current outer wall image and the matching image conforming to the environmental data of the day are extracted respectively, and the texture features of the normal image conforming to the environmental data of the day are extracted;

[0072] Based on the extracted texture features, the feature mode of the matching image conforming to the environmental data of the day and the feature mode of the normal image conforming to the environmental data of the day are calculated;

[0073] The first difference value of the texture features of the current outer wall image and the feature mode of the matching image is calculated, and the second difference value of the texture features of the current outer wall image and the feature mode of the normal image is calculated;

[0074] If the first difference value is less than the second difference value, and the first difference value is less than the hollow difference value threshold, and the texture features of the current outer wall image meet the conditional constraint based on the feature mode of the hollow image in the pre-constructed outer wall image library, the initial judgment result that the current outer wall has hollow is generated.

[0075] For the newly collected outer wall infrared image, the texture features and gray features are extracted by using image processing algorithm, the energy, entropy, and contrast texture parameters are calculated through the gray level co-occurrence matrix, the quantitative image is selected from the similar image set, the feature parameters are extracted, and the features of the newly collected image and the similar image are compared one by one, the difference value between the features is calculated, and the similarity of the features is measured.

[0076] Based on the rules explored in the comparison and analysis of data A and data B, the feature difference between the newly collected image and the similar image is judged; if the feature difference between the newly collected image and the similar image with hollow is small, and the feature mode of the hollow image conforms to the known rule, it is preliminarily judged that the current wall has hollow; if the feature is close to the feature mode of the normal image, it is preliminarily judged that the current wall does not have hollow.

[0077] The energy, entropy, and contrast texture parameters are calculated through the gray level co-occurrence matrix, and the image features are quantitatively described from multiple dimensions. Compared with single feature analysis, the comprehensive consideration of multiple features can more comprehensively reflect the condition of the outer wall surface.

[0078] For the newly collected outer wall infrared image, the texture features and gray features are extracted by using image processing algorithm, the energy, entropy, and contrast texture parameters are calculated through the gray level co-occurrence matrix, and the specific calculation method is:

[0079] Energy is used to measure the uniformity of the image gray scale distribution, for each element in the gray scale co-occurrence matrix, the element value is squared, and the results of all elements after squaring are added, the higher the energy value, the more concentrated the image gray scale distribution, the less the gray scale changes, and the smoother and more uniform the image looks; on the contrary, the lower the energy value, the more dispersed the gray scale distribution;

[0080] Entropy is used to measure the complexity and randomness of the image gray scale distribution, each element of the gray scale co-occurrence matrix is checked, if the element value is not zero, the element value is multiplied by the inverse of its logarithm with base 2; if the element value is zero, the calculation result is zero, the results of all elements after the above calculation are added to obtain the entropy value, the greater the entropy value, the more complex and disordered the distribution of gray scale in the image, the more details and changes; the smaller the entropy value, the simpler and more regular the image gray scale distribution;

[0081] Contrast is used to reflect the degree of local gray scale change in the image, the higher the contrast value, the more obvious the difference between light and dark areas in the image, and the more prominent the edges and details of the image; the lower the contrast value, the more gentle the gray scale change of the image, and the more blurred in vision.

[0082] Let the gray scale co-occurrence matrix be P(i,j), where i and j represent the gray scale level, and the matrix size is N x N (N is the number of gray scale levels);

[0083] Energy is used to measure the uniformity of the image gray scale distribution, the calculation formula is:

[0084]

[0085] The higher the energy value, the more concentrated the image gray scale distribution, the less the gray scale changes, and the smoother and more uniform the image looks; on the contrary, the lower the energy value, the more dispersed the gray scale distribution;

[0086] Entropy is used to measure the complexity and randomness of the image gray scale distribution, the calculation formula is:

[0087]

[0088] When P(i,j) = 0, P(i,j) log2 P(i,j) = 0 is specified, the greater the entropy value, the more complex and disordered the distribution of gray scale in the image, the more details and changes; the smaller the entropy value, the simpler and more regular the image gray scale distribution;

[0089] Contrast is used to reflect the degree of local gray scale change in the image, the calculation formula is:

[0090]

[0091] Where n = |ij|, that is, the difference between gray levels i and j. The higher the contrast value, the more obvious the difference between bright and dark areas in the image, and the more prominent the edges and details of the image. The lower the contrast value, the more gradual the gray level change of the image, and the more blurred it appears visually.

[0092] Let the set of quantitative images selected from the set of similar images be S = {S1, S2, ..., S...} n The newly acquired image is I. new ;

[0093] Extract newly acquired image I new The eigenvectors of are denoted as .

[0094] For each image S in the set of similar images S i Extract the feature vector from (i = 1, 2, ..., n), denoted as

[0095] The difference between the feature vector of the newly acquired image and the feature vector of similar images is calculated using the Euclidean distance formula. Let the difference value be D. i The calculation formula is:

[0096]

[0097] Where m is the dimension of the feature vector, F new,j Represents the feature vector of the newly acquired image The j-th eigenvalue, Represents similar images S i Feature vector The j-th eigenvalue;

[0098] Difference value D i The smaller the value, the better the newly acquired image I. new Similar image S i The more similar their features;

[0099] Suppose that a set of similar images containing hollow areas is known to exist. Its characteristic pattern is denoted as The characteristic patterns of a normal image are denoted as

[0100] Calculate the feature vector of the newly acquired image Hollow image feature patterns Difference D hollow and normal image feature patterns Difference D normal The Euclidean distance method is also used for calculation:

[0101]

[0102] Rule: if D hollOw <D normal and D hollow is less than a certain threshold value hollow , and The feature mode of the known rule is satisfied. The condition judgment function is:

[0103]

[0104] If the condition is satisfied, return true, otherwise return false. The preliminary judgment of the current wall is that there is a hollow drum, that is, the judgment result exists hollow drum, and true; there is no hollow drum, and other conditions:

[0105]

[0106] Through the formula and the judgment condition, based on the feature comparison of similar images and the known rule, whether the wall corresponding to the newly collected image exists hollow drum is preliminarily judged.

[0107] The present application adopts the Euclidean distance formula to calculate the difference value between the feature vectors of the newly collected image and the similar image, and quantitatively processes the image feature difference. In actual detection, the texture, gray scale and other features of the image are relatively abstract. The Euclidean distance formula can convert these abstract features into specific numerical values, and intuitively reflect the similarity between the features. When there is a slight difference between the texture details and the gray scale distribution of the newly collected image and the similar image, the difference value can be accurately obtained through the Euclidean distance calculation, which provides accurate data support for subsequent judgment. Compared with subjective judgment or non-quantitative comparison method, the accuracy of the detection result is greatly improved. The differences between the feature vectors of the newly collected image and the feature mode of the hollow drum image and the feature mode of the normal image are calculated respectively, forming a two-way comparison mechanism. This way avoids the risk of misjudgment when only compared with a single mode.

[0108] In an optional embodiment of the present application, the current outer wall image and the environmental data of the day are collected, and the matching image conforming to the environmental data of the day is searched from the pre-constructed outer wall image library, comprising:

[0109] Collect the current outer wall image and the weather and temperature of the day;

[0110] Take the weather and temperature of the day as the search condition, and screen whether there is the same weather and temperature as the weather and temperature of the day in the pre-constructed outer wall image library;

[0111] If there is the same weather and temperature as the weather and temperature of the day, the image associated with the same weather and temperature is taken as the matching image conforming to the environmental data of the day;

[0112] If there is no weather and temperature that is exactly the same as the weather and temperature of the day, the image associated with the closest weather and temperature is selected as the matching image that conforms to the environmental data of the day.

[0113] In this embodiment, the recorded temperature and weather of the day are taken as the retrieval conditions to screen in the constructed image library; the image that completely matches the environmental conditions of the day is preferentially searched, and if there is no completely matching image, the image under the closest temperature and weather conditions is searched to determine the similar image set.

[0114] The present application can fully consider the interference of environmental factors on the detection result through multi-source data acquisition, so that the subsequent analysis is established on the data basis that is more in line with the actual scene, and the accuracy and reliability of the detection are greatly improved; the completely matching image is preferentially searched, and if there is no completely matching image, the image under the closest condition is selected to quickly determine the similar image set, which avoids comparing all images in the image library one by one, greatly reduces the data processing amount, and significantly improves the detection efficiency; at the same time, the image under the similar environment is taken as the reference, the features of the newly collected image can be more reasonably evaluated, and the detection result is more persuasive.

[0115] In an optional embodiment of the present application, the knocking sound frequency of the current outer wall is collected to verify whether the initial judgment result is accurate, and the outer wall detection result is generated, including:

[0116] The knocking sound signal at the hollow position of the outer wall marked by the initial judgment result is collected, the knocking sound signal is preprocessed, and the frequency feature is extracted through spectrum analysis;

[0117] The similarity between the frequency feature and the hollow sample in the hollow sound frequency database is calculated.

[0118] If the initial judgment result is that the outer wall is hollow, and the similarity is higher than the preset similarity threshold, the outer wall detection result that the current outer wall is hollow is generated.

[0119] In the field of modern building detection, in order to efficiently and accurately complete the hollow detection of the outer wall, the detection scheme of unmanned aerial vehicle carrying special equipment is innovatively adopted. In the equipment selection link, according to the height, structural complexity and detection environment requirements of the building outer wall, the appropriate unmanned aerial vehicle model is selected. The unmanned aerial vehicle needs to have long endurance to meet the flight time requirements of large-area outer wall detection. At the same time, it needs to have high stability to maintain hovering attitude in complex airflow environment and ensure the normal work of the detection equipment. On the unmanned aerial vehicle platform, the customized knocking hammer and high-sensitivity sound sensor are carefully installed. The force and frequency of the knocking hammer are accurately adjusted to ensure that effective knocking is generated without damaging the wall surface, and to make the hollow area and the normal area produce high distinguishable sound difference. The sound sensor can accurately capture the subtle sound signals generated by the wall after knocking.

[0120] After completing the equipment installation, based on the suspected hollow positions marked by other detection methods in the early stage, the flight path of the unmanned aerial vehicle is planned by using professional path planning software. The planning process fully considers the distribution of obstacles such as building outline, windows and balconies, and generates a safe and efficient route. The unmanned aerial vehicle flies to each detection point according to the preset path, relies on advanced positioning system and hovering control technology to hover stably above the detection point. At this time, the remote control system sends instructions to control the knocking hammer to knock the wall with constant force and frequency. The sound signals generated by the knocking are quickly collected by the sound sensor. After the sensor converts the analog signal into a digital signal, it is transmitted in real time to the ground terminal equipment through the wireless transmission module.

[0121] After the ground terminal receives the sound signal, it immediately performs preprocessing, which includes signal filtering to remove irrelevant signals such as environmental noise and electromagnetic interference; signal amplification to enhance the strength of weak valid signals and ensure the accuracy of subsequent analysis. The preprocessed signal enters the frequency spectrum analysis stage, where fast Fourier transform (FFT) and other algorithms are used to convert time domain signals into frequency domain signals and extract frequency characteristics. These frequency characteristics contain key information about the wall material, structure and whether there is hollow. The frequency characteristics are compared with the hollow sound frequency database to calculate the similarity, and the results are combined with the preliminary detection results for judgment. If the sound frequency is similar to the hollow sample and consistent with the preliminary hollow result, it is confirmed as hollow. Otherwise, further analysis is needed. If the initial judgment result is that the outer wall has hollow, and the similarity is not higher than the preset similarity threshold, record both judgment results and remind the engineer to verify. If the initial judgment result is that the outer wall has not hollow, and the similarity is higher than the preset similarity threshold, record both judgment results and remind the engineer to verify. If the initial judgment result is that the outer wall has not hollow, and the similarity is not higher than the preset similarity threshold, generate the outer wall detection result that the current outer wall has not hollow. Form a detection report.

[0122] The present application guarantees reliable results through infrared thermal imaging and unmanned aerial vehicle sound wave double verification, ensures personnel safety through non-contact detection, reduces operation cost through remote control of the unmanned aerial vehicle, avoids the influence of meteorological conditions, and reduces the wall temperature difference, thereby avoiding the situation of reduced detection effect.

[0123] It should be noted that, for the method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the embodiments of the present application are not limited by the action sequence described, because according to the embodiments of the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions involved are not necessarily necessary for the embodiments of the present application.

[0124] Referring to Figure 2 , a structural block diagram of a kind of outer wall hollow detection system based on machine learning provided in the embodiments of the present application is shown, specifically can include following module:

[0125] Environment data image matching module 201 is used to collect current outer wall image and daily environment data, and find matching image conforming to daily environment data from pre-constructed outer wall image library; Pre-constructed outer wall image library stores hollow image under complex environment, hollow image under normal environment; Environment data includes weather and temperature;

[0126] Outer wall hollow initial judgment module 202 is used to compare features of current outer wall image and matching image conforming to daily environment data, and analyze whether current outer wall occurs hollow by combining pre-constructed hollow judgment rule, generate initial judgment result; Pre-constructed hollow judgment rule is constructed by analyzing outer wall hollow image feature mode and feature change rule based on pre-constructed outer wall image library;

[0127] Outer wall hollow verification module 203 is used to collect the knocking sound frequency of current outer wall, verify whether initial judgment result is accurate, generate outer wall detection result.

[0128] In an optional embodiment of the present application, the system further includes an outer wall image library construction module, which is used to:

[0129] Numbering partitioning is performed on the outer wall of known hollow area;

[0130] Complex environment data and hollow images and normal images under complex environment, and normal environment data and hollow images and normal images under normal environment are collected for hollow area and normal area of outer wall;

[0131] Sort and classify the collected images, and associate each hollow drum image and normal image with corresponding weather and temperature data.

[0132] In an optional embodiment of the present application, the system further comprises an analysis module, which is configured to:

[0133] extract texture features of the hollow drum image and the normal image in the complex environment, respectively;

[0134] calculate the difference between the texture features of the hollow drum image and the normal image in the complex environment, form a feature difference matrix, and convert the feature difference matrix into a heat map;

[0135] group the hollow drum images according to weather and temperature, and calculate a hollow drum image feature mode; the feature mode is a statistical value of the texture features of the hollow drum image;

[0136] analyze the feature change rule based on the heat map and the hollow drum image feature mode, and construct a hollow drum judgment rule based on the hollow drum image feature mode and the feature change rule.

[0137] In an optional embodiment of the present application,

[0138] the feature change rule is that, compared with the normal environment, the dry and hot environment increases the feature difference between the hollow drum region and the normal region; the wet and cold environment reduces the feature difference between the hollow drum region and the normal region; the normal environment includes sunny and normal temperature; the dry and hot environment includes sunny and high temperature; and the wet and cold environment includes rainy, overcast and low temperature;

[0139] the hollow drum judgment rule is that, if the feature of the current external wall image is small compared with the feature of the matching image that meets the environmental data of the day, and at the same time meets the feature mode condition constraint of the hollow drum image in the feature change rule, it is determined that the current external wall has hollow drum.

[0140] In an optional embodiment of the present application, the external wall hollow drum initial judgment module comprises:

[0141] a texture feature extraction submodule, configured to extract texture features of the current external wall image and the matching image that meets the environmental data of the day, respectively, and extract texture features of the normal image that meets the environmental data of the day;

[0142] a feature mode calculation submodule, configured to calculate the feature mode of the matching image that meets the environmental data of the day and the feature mode of the normal image that meets the environmental data of the day based on the extracted texture features;

[0143] a difference value calculation submodule, configured to calculate a first difference value between the texture features of the current external wall image and the feature mode of the matching image, and calculate a second difference value between the texture features of the current external wall image and the feature mode of the normal image;

[0144] The outer wall hollowing initial judgment submodule is configured to generate an initial judgment result that the current outer wall is hollowed if the first difference value is smaller than the second difference value, the first difference value is smaller than the hollowing difference value threshold, and the texture feature of the current outer wall image meets the conditional constraint based on the feature mode constructed based on the pre-constructed outer wall image library of the hollowing image.

[0145] In an optional embodiment of the present application, the environmental data image matching module comprises:

[0146] The acquisition submodule is configured to acquire the current outer wall image, the weather and the temperature of the day;

[0147] The screening submodule is configured to screen whether there is the same weather and temperature as the weather and the temperature of the day in the pre-constructed outer wall image library by taking the weather and the temperature of the day as the retrieval condition;

[0148] The first matching submodule is configured to take the image associated with the same weather and temperature as the matching image conforming to the environmental data of the day if there is the same weather and temperature as the weather and the temperature of the day;

[0149] The second matching submodule is configured to take the image associated with the closest weather and temperature as the matching image conforming to the environmental data of the day if there is no weather and temperature completely same as the weather and the temperature of the day.

[0150] In an optional embodiment of the present application, the texture feature comprises energy, entropy and contrast.

[0151] In an optional embodiment of the present application, the system further comprises:

[0152] The gray level co-occurrence matrix establishing module is configured to establish the gray level co-occurrence matrix of each image for the current outer wall image, the matching image conforming to the environmental data of the day and the normal image conforming to the environmental data of the day, respectively;

[0153] The texture feature calculating module is configured to calculate the energy, the entropy and the contrast through the gray level co-occurrence matrix of each image.

[0154] In an optional embodiment of the present application, the outer wall hollowing verification module comprises:

[0155] The sound frequency feature extraction submodule is configured to acquire the knocking sound signal at the hollowed position of the outer wall marked by the initial judgment result, pre-process the knocking sound signal and extract the frequency feature through the spectrum analysis;

[0156] The sound frequency feature similarity calculating submodule is configured to calculate the similarity between the frequency feature and the hollowing sample in the hollowing sound frequency database;

[0157] The outer wall detection result generation submodule is configured to generate an outer wall detection result that the current outer wall is hollow if the initial judgment result is that the outer wall is hollow and the similarity is higher than the preset similarity threshold.

[0158] For the system embodiments, the description is relatively simple because the system embodiments are basically similar to the method embodiments. For the relevant parts, refer to the description of the method embodiments.

[0159] It should be noted that, in this document, the terms such as first and second are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0160] Each of the embodiments in the specification is described in a relevant manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the system embodiments, the description is relatively simple because the system embodiments are basically similar to the method embodiments. For the relevant parts, refer to the description of the method embodiments.

[0161] The above only describes the preferred embodiments of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting hollowing of an exterior wall based on machine learning, characterized by, The method comprises: Collecting a current outer wall image and daily environment data, and searching for a matching image conforming to the daily environment data from a pre-constructed outer wall image library; the pre-constructed outer wall image library stores a hollow drum image under a complex environment and a hollow drum image under a normal environment; the environment data comprises weather and temperature; Comparing features of the current outer wall image and the matching image conforming to the daily environment data, and analyzing whether the current outer wall has a hollow drum in combination with a pre-constructed hollow drum judgment rule to generate an initial judgment result; the pre-constructed hollow drum judgment rule is constructed by analyzing a hollow drum image feature mode and a feature change rule based on the pre-constructed outer wall image library; Collecting a knocking sound frequency of the current outer wall to verify whether the initial judgment result is accurate to generate an outer wall detection result; Analyzing a hollow drum image feature mode and a feature change rule based on the outer wall image library to construct a hollow drum judgment rule, comprising: Respectively extracting texture features of the complex environment hollow drum image and the normal environment hollow drum image; Calculating differences between the texture features of the complex environment hollow drum image and the normal environment hollow drum image to form a feature difference matrix, and converting the feature difference matrix into a heat map; Grouping the hollow drum images according to weather and temperature, and calculating a hollow drum image feature mode; the feature mode is a statistical value of the texture features of the hollow drum images; Analyzing a feature change rule according to the heat map and the hollow drum image feature mode, and constructing a hollow drum judgment rule based on the hollow drum image feature mode and the feature change rule; The feature change rule is that, compared with a normal environment, a dry and hot environment increases feature differences between a hollow drum region and a normal region; a wet and cold environment reduces the feature differences between the hollow drum region and the normal region; the normal environment comprises a sunny day and a normal temperature; the dry and hot environment comprises a sunny day and a high temperature; and the wet and cold environment comprises a rainy day, a cloudy day and a low temperature; The hollow drum judgment rule is that, if a feature difference between the current outer wall image and the matching image conforming to the daily environment data is small, and the feature difference meets a feature mode condition constraint of the hollow drum image in the feature change rule, it is determined that the current outer wall has a hollow drum.

2. The method of claim 1, wherein, Constructing an outer wall image library, comprising: Numbering and partitioning an outer wall of a known hollow drum region; Collecting complex environment data, a hollow drum image and a normal image under a complex environment, and normal environment data, a hollow drum image and a normal image under a normal environment for the hollow drum region and the normal region of the outer wall; Arranging and classifying the collected images to associate each hollow drum image and normal image with corresponding weather and temperature data.

3. The method of claim 1, wherein, Comparing features of the current outer wall image and the matching image conforming to the daily environment data, and analyzing whether the current outer wall has a hollow drum in combination with a pre-constructed hollow drum judgment rule to generate an initial judgment result, comprising: Respectively extracting texture features of the current outer wall image and the matching image conforming to the daily environment data, and extracting texture features of a normal image conforming to the daily environment data; Calculating a feature mode of the matching image conforming to the daily environment data and a feature mode of the normal image conforming to the daily environment data based on the extracted texture features; a first difference value between the texture feature of the current outer wall image and the feature mode of the matching image, and a second difference value between the texture feature of the current outer wall image and the feature mode of the normal image; if the first difference value is less than the second difference value, and the first difference value is less than the hollow difference value threshold, and the texture feature of the current outer wall image meets the conditional constraint constructed based on the feature mode of the hollow image in the pre-constructed outer wall image library, an initial judgment result that the current outer wall has hollow is generated.

4. The method of claim 1, wherein, collecting the current outer wall image and the environmental data of the day, and searching for a matching image meeting the environmental data of the day from the pre-constructed outer wall image library, comprising: collecting the current outer wall image and the weather and temperature of the day; using the weather and temperature of the day as a search condition, screening whether there is the same weather and temperature in the pre-constructed outer wall image library; if there is the same weather and temperature, the image associated with the same weather and temperature is taken as the matching image meeting the environmental data of the day; if there is no weather and temperature completely same as the weather and temperature of the day, the image associated with the closest weather and temperature is taken as the matching image meeting the environmental data of the day.

5. The method of claim 1, wherein, The texture feature includes energy, entropy and contrast.

6. The method of claim 5, wherein, The method further comprises: establishing a gray level co-occurrence matrix of each image respectively for the current outer wall image, the matching image meeting the environmental data of the day and the normal image meeting the environmental data of the day; calculating the energy, entropy and contrast through the gray level co-occurrence matrix of each image.

7. The method of claim 1, wherein, collecting the knocking sound frequency of the current outer wall, verifying whether the initial judgment result is accurate, and generating an outer wall detection result, comprising: collecting the knocking sound signal at the hollow position of the outer wall marked by the initial judgment result, pre-processing the knocking sound signal and extracting the frequency feature through frequency spectrum analysis; calculating the similarity between the frequency feature and the hollow sample in the hollow sound frequency database; if the initial judgment result is that the outer wall has hollow, and the similarity is higher than the preset similarity threshold, an outer wall detection result that the current outer wall has hollow is generated. 8.A machine learning based external wall hollow detection system, characterized in that, The system comprises: an environmental data image matching module for collecting the current outer wall image and the environmental data of the day, and searching for a matching image meeting the environmental data of the day from the pre-constructed outer wall image library; the pre-constructed outer wall image library stores the hollow images under complex environment and the hollow images under normal environment; the environmental data includes weather and temperature; an outer wall hollow initial judgment module for comparing the features of the current outer wall image and the matching image meeting the environmental data of the day, and analyzing whether the current outer wall has hollow in combination with the pre-constructed hollow judgment rule, and generating an initial judgment result; the pre-constructed hollow judgment rule is constructed based on the pre-constructed outer wall image library by analyzing the feature mode and the feature change rule of the outer wall hollow image; an outer wall hollow verification module for collecting the knocking sound frequency of the current outer wall, verifying whether the initial judgment result is accurate, and generating an outer wall detection result; The system further comprises an analysis module for: extracting the texture feature of the complex environment hollow image and the normal environment hollow image respectively; Differences between the texture features of the complex environment hollow image and the texture features of the normal environment hollow image are calculated to form a feature difference matrix, and the feature difference matrix is converted into a heat map; The hollow images are grouped according to weather and temperature, and a hollow image feature mode is calculated; the feature mode is a statistical value of the texture features of the hollow images; Characteristic variation laws are analyzed according to the heat map and the hollow image feature mode, and a hollow judgment rule is constructed based on the hollow image feature mode and the characteristic variation laws of the external wall; The characteristic variation laws are that, compared with a normal environment, the feature difference between the hollow region and the normal region is increased in a dry and hot environment, and the feature difference between the hollow region and the normal region is reduced in a wet and cold environment; the normal environment includes sunny and regular temperature; the dry and hot environment includes sunny and high temperature; and the wet and cold environment includes rainy, overcast and low temperature; The hollow judgment rule is that, if the feature difference between the current external wall image and the matching image in accordance with the current environmental data is small, and the feature mode of the hollow image in the characteristic variation laws is satisfied, then it is determined that the current external wall has hollowing.

Images