An AI-based method for detecting tunnel surface defects

Abstract

This application discloses an artificial intelligence-based method for detecting tunnel exterior defects, belonging to the field of data processing technology. The method includes acquiring general defect data, tunnel features, and engineering design parameters of the target tunnel for different tunnels to obtain a basic defect database model; acquiring specific defect data for different tunnels, identifying texture, shape, and color features of the specific defect data to obtain a specific defect database model; generating a drone's identification route based on the engineering design parameters within the target tunnel; acquiring visible light and infrared thermal images of the target tunnel's exterior and identifying defect points; identifying defect points and performing correlation analysis to obtain composite defect risks; constructing a three-dimensional point cloud model of the target tunnel, building a three-dimensional defect map, and formulating a maintenance plan to obtain a maintenance treatment plan. This application improves the accuracy and efficiency of detecting exterior defects inside tunnels.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an artificial intelligence-based method for detecting tunnel surface defects. Background Technology

[0002] Tunnels are typically found in mountainous areas, undersea tunnels, and other similar locations to construct roads, facilitating travel for cars and trains. Therefore, tunnels play a vital role in transportation, and their safety is paramount. After construction, tunnels require continuous inspection and maintenance to ensure the safety of vehicles traveling within them.

[0003] In existing technologies, the inspection of tunnel exterior defects is generally conducted manually. However, due to the considerable height of tunnels, the accuracy of visual inspection is limited, and insufficient lighting conditions inside the tunnel make it difficult to clearly identify defects in areas that are difficult to see with the naked eye. This method not only consumes a large amount of manpower but also results in insufficient accuracy and unreliable results. Furthermore, manual inspection relies heavily on the experience of the staff, especially when inspecting minute details, easily overlooking deeper causes behind minor defects, leading to subsequent losses. Therefore, how to efficiently and accurately inspect the exterior of tunnels has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based method for detecting tunnel surface defects, in order to solve the problems mentioned in the background art.

[0005] This application provides an artificial intelligence-based method for detecting tunnel appearance defects, the method comprising: Acquire common defect data, tunnel characteristics, and engineering design parameters of the target tunnel from different tunnels, construct a defect database model, and perform transfer learning on the defect database model based on the engineering design parameters to obtain a basic defect database model. Acquire special disease data from different tunnels, identify the texture, shape and color features of the special disease data, and generate diverse synthetic disease data based on conditional generative adversarial network. Combine the special disease data and the synthetic disease data to train the model and obtain a special disease database model. Based on the engineering design parameters within the target tunnel, the internal structure data of the tunnel is obtained, and the identification working route of the UAV is generated according to the internal structure data. The drone conducts inspections according to the identified work route, and simultaneously collects visible light images and infrared thermal images of the target tunnel's exterior, and identifies the defects. Based on the basic disease database model and the special disease database model, the disease points are identified and correlation analysis is performed to obtain the risk of compound diseases. Based on the internal structure data and the infrared thermal imaging image, a three-dimensional point cloud model of the target tunnel is constructed. Combining the three-dimensional point cloud model and the composite disease risk, a three-dimensional disease map is constructed. Based on the three-dimensional disease map, a maintenance plan is formulated to obtain a maintenance treatment plan.

[0006] Preferably, the steps of constructing a disease database model and performing transfer learning on the disease database model based on the engineering design parameters to obtain a basic disease database model are as follows: Based on the tunnel characteristics, the service life, design purpose, usage scenarios, and appearance of different tunnels are obtained. Based on the service life, design purpose, usage scenario, and appearance shape, a database model framework for multi-source data fusion is generated; Based on the general disease data, disease identification is performed on the general disease data to obtain multiple general disease types, as well as the disease severity value and disease treatment plan corresponding to each general disease type; Based on the general disease type, the disease severity value, and the disease treatment plan, generate the content of a database model that integrates multi-source data, and construct a disease database model by combining the database model framework and the database model content. Based on the engineering design parameters of the target tunnel, the disease database model is migrated, and the migrated data is re-integrated to generate a basic disease database model.

[0007] Preferably, the steps of identifying the texture, shape, and color features of the specific disease data, generating diverse synthetic disease data based on a conditional generative adversarial network, and training a model by combining the specific disease data and the synthetic disease data to obtain a specific disease database model are as follows: Feature recognition is performed on the special disease data to obtain the texture features, shape features and color features of the special disease data respectively, and the texture features, shape features and color features are combined to generate a data synthesis blueprint; Based on the data synthesis blueprint, an adversarial network is generated according to the conditions to produce diverse synthetic disease data; The synthetic disease data and the special disease data are stored in the same set and shuffled to generate a model training package; An initial special disease database model is constructed, and the initial special disease database model is trained using the model training package until the initial special disease database model can identify all special disease data in the model training package, thus obtaining the special disease database model.

[0008] Preferably, the step of obtaining the internal structure data of the tunnel based on the engineering design parameters within the target tunnel, and generating the UAV's identification route based on the internal structure data, specifically includes: Based on the engineering design parameters within the target tunnel, parameter extraction is performed on the engineering design parameters to obtain tunnel structure data and external construction data. Based on the tunnel construction data and the exterior construction data, the internal shape and size parameters of the tunnel are obtained, and the internal structure data is obtained based on the shape and size parameters. The drone's single inspection range and fuselage size parameters are obtained. Combined with the internal structure data, the single inspection range, and the fuselage size parameters, the drone's initial working route is generated. The frequency of vehicle traffic within the tunnel is obtained, and the initial working route is adjusted based on the vehicle traffic frequency to generate an identification working route for the drone.

[0009] Preferably, the step of the drone conducting inspections according to the identified work route, simultaneously acquiring visible light images and infrared thermal images of the target tunnel's exterior, and identifying the defects, specifically includes: The drone conducts inspections according to the identified working route, and collects images and thermal imaging data of the inner wall of the target tunnel, respectively, to obtain visible light images and infrared thermal imaging images. Based on the visible light image, the brightness and contrast of different regions in the visible light image are identified to obtain underexposed and overexposed regions. The underexposed areas are locally enhanced, and the overexposed areas are locally suppressed to obtain a visible light image of the target. Image recognition is performed on the target visible light image to identify abnormal image regions present in the image; Based on the infrared thermal imaging image, areas with temperature differences are identified to obtain abnormal temperature areas; By combining the abnormal image area and the abnormal temperature area, the disease points are obtained.

[0010] Preferably, the step of performing image recognition on the target visible light image to obtain abnormal image regions in the image specifically includes: Obtain dust image samples, water stain image samples, and old trace image samples inside the tunnel, and extract the dust features of the dust image samples, the water stain features of the water stain image samples, and the old trace features of the old trace image samples, respectively. By combining the dust characteristics, the water stain characteristics, and the old trace characteristics, environmental interference factor characteristics are generated; The visible light image is filtered for interference factors based on the environmental interference factor characteristics to obtain a clean target visible light image. The processed target visible light image is then used for image recognition to identify abnormal image areas.

[0011] Preferably, after the step of simultaneously acquiring visible light images and infrared thermal images of the target tunnel's interior and exterior, and identifying the defects, the method further includes: Based on the disease points, the actual size and disease type of the disease points are identified, and a risk assessment is performed based on the actual size and disease type to obtain an initial risk value; Determine whether the initial risk value is less than the preset standard risk value. If the initial risk value is less than the standard risk value, then track and monitor the disease point to obtain disease change data over a period of time. Based on the disease change data, the disease evolution trend of the disease point is obtained, and the risk change degree label is obtained by extrapolation based on the disease evolution trend. The risk change degree label is then affixed to the location of the disease point.

[0012] Preferably, the step of identifying the disease points and performing correlation analysis based on the basic disease database model and the special disease database model to obtain the risk of compound diseases specifically includes: Based on the basic disease database model and the special disease database model, the disease points are identified to obtain the current disease type and current disease range of the disease points; Based on the current disease type and the current disease range, data association is performed in the basic disease database model and the special disease data model to obtain target associated data; Disease extraction is performed on the target associated data to obtain associated disease data, and the associated disease data is matched according to the current disease type to obtain multiple matching values; Determine whether the matching value exceeds a preset standard value. If the matching value exceeds the standard value, mark the associated disease data as candidate associated disease data. The explicit features of the candidate associated disease data are extracted, and the explicit features are compared and screened with the visible light image and the infrared thermal image to obtain the target associated disease data; By combining the current disease type, the current disease range, and the target associated disease data, composite disease parameters are obtained. A risk assessment is then performed on the composite disease parameters to obtain the composite disease risk.

[0013] Preferably, the steps of constructing a 3D point cloud model of the target tunnel, combining the 3D point cloud model with the composite defect risk to construct a 3D defect map, and formulating a maintenance plan based on the 3D defect map to obtain the maintenance treatment plan are as follows: Based on the internal structure data and the infrared thermal imaging image, the external structure of the tunnel's internal surface is scanned to generate point cloud data of the tunnel. Based on the point cloud data, the interior of the target tunnel is modeled to generate a three-dimensional point cloud model of the target tunnel. The composite disease parameters are reproduced in the three-dimensional point cloud model to obtain the three-dimensional point cloud model of the target disease. By combining the target disease 3D point cloud model and the composite disease risk, a 3D disease map is constructed, and the cause of the disease is determined based on the 3D disease map. Based on the cause of the disease, a maintenance plan is formulated to obtain a maintenance treatment plan.

[0014] In summary, this application includes at least one of the following beneficial technical effects: By acquiring general defect data, tunnel characteristics, and specific defect data from different tunnels, as well as the engineering design parameters of the target tunnel, basic defect database models and specific defect database models are constructed. Based on the engineering design parameters, internal tunnel structure data is obtained. Then, a drone identification route is generated based on this internal structure data. The drone performs inspections according to the identification route, continuously collecting visible light and infrared thermal images of the tunnel's inner wall to identify defect points. These defect points are then identified using the basic and specific defect database models, followed by correlation analysis to determine the risk of complex defects. Next, a 3D point cloud model of the target tunnel is constructed based on the internal structure data and infrared thermal images. The parameters of the complex defects are reproduced within this 3D point cloud model to obtain a 3D point cloud model of the target defects. Simultaneously, combined with the risk of complex defects, a 3D defect map is generated. The 3D defect map is then analyzed to determine the causes of the defects, and maintenance and treatment plans are developed based on these causes. This improves the accuracy and efficiency of detecting defects in the tunnel's internal appearance. Attached Figure Description

[0015] Figure 1This application provides a step flow of an artificial intelligence-based method for detecting tunnel appearance defects. Detailed Implementation

[0016] The following combination Figure 1 This application will be described in further detail, but the embodiments of the present invention are not limited thereto.

[0017] This application discloses an artificial intelligence-based method for detecting tunnel appearance defects.

[0018] In this embodiment, an artificial intelligence-based method for detecting tunnel appearance defects includes: S100: Obtain common defect data, tunnel characteristics and engineering design parameters of the target tunnel from different tunnels, construct a defect database model, and perform transfer learning on the defect database model based on the engineering design parameters to obtain a basic defect database model. S200: Acquire special disease data from different tunnels, identify the texture, shape and color features of the special disease data, and generate diverse synthetic disease data based on conditional generative adversarial network. Combine the special disease data and the synthetic disease data to train the model and obtain the special disease database model. S300: Based on the engineering design parameters within the target tunnel, obtain the internal structure data of the tunnel, and generate the identification working route of the UAV based on the internal structure data; S400: The UAV conducts inspections according to the identified work route, and simultaneously collects visible light images and infrared thermal images of the target tunnel's exterior, and identifies the defects. S500: Based on the basic disease database model and the special disease database model, disease points are identified and correlation analysis is performed to obtain the risk of compound diseases; S600: Based on internal structural data and infrared thermal imaging images, a three-dimensional point cloud model of the target tunnel is constructed. Combining the three-dimensional point cloud model with the risk of complex defects, a three-dimensional defect map is constructed. Based on the three-dimensional defect map, a maintenance plan is formulated to obtain the maintenance treatment plan.

[0019] It should be noted that the above process is only the basic steps of this embodiment. In the specific implementation process, some steps may be added, reduced or modified appropriately without affecting the overall implementation effect.

[0020] The steps for constructing a disease database model and performing transfer learning on the disease database model based on engineering design parameters to obtain a basic disease database model are as follows: Based on tunnel characteristics, the service life, design purpose, usage scenarios, and appearance of different tunnels are obtained; Based on the service life, design purpose, usage scenario, and appearance, a database model framework for multi-source data fusion is generated. Based on common disease data, disease identification is performed on the common disease data to obtain various common disease types, as well as the disease severity value and disease treatment plan for each common disease type; Based on common disease types, disease severity values, and disease treatment plans, generate database model content for multi-source data fusion, and construct a disease database model by combining the database model framework and database model content. Based on the engineering design parameters of the target tunnel, the disease database model is migrated, and the migrated data is re-integrated to generate a basic disease database model.

[0021] In this application, a traffic tunnel connecting City A and City B is used as an example. Built five years ago, this tunnel is designed as a two-way four-lane highway tunnel for daily vehicular traffic, and has a horseshoe-shaped arch structure. First, based on the tunnel's service life, design purpose, usage scenario, and appearance, a multi-source data fusion database model framework is generated. This framework includes the tunnel's basic information, structural parameters, and environmental characteristics. Next, common defects data for the tunnel over the past five years are collected, including common defect types such as cracks, leakage, and spalling. By identifying these common defects, the severity of cracks is determined to be moderate, leakage severity to be slight, and spalling severity to be severe. Simultaneously, corresponding treatment plans are developed for each defect type, such as grouting repair for cracks, waterproofing for leakage, and surface repair for spalling. Based on these common defect types, defect severity values, and treatment plans, the content of the multi-source data fusion database model is generated. Finally, combining the database model framework and content, a complete defect database model is constructed. Based on the engineering design parameters of the target tunnel, such as tunnel length, cross-sectional dimensions, and material type, the defect database model is migrated and re-integrated to generate a basic defect database model suitable for the tunnel.

[0022] The steps involved in identifying texture, shape, and color features of specific disease data, generating diverse synthetic disease data using a conditional generative adversarial network, and training a model by combining the specific and synthetic disease data to obtain a specific disease database model are as follows: Feature identification is performed on special disease data to obtain the texture features, shape features and color features of the special disease data respectively, and the texture features, shape features and color features are combined to generate a data synthesis blueprint; Based on the data synthesis blueprint, an adversarial network is generated according to the conditions to produce diverse synthetic disease data; Synthetic disease data and special disease data are stored in the same set and shuffled to generate a model training package; An initial special disease database model is constructed, and the initial special disease database model is trained using a model training package until the initial special disease database model can identify all special disease data in the model training package, thus obtaining the special disease database model.

[0023] In this application, taking a traffic tunnel connecting City A and City B as an example, data on specific defects that appeared in the tunnel over the past five years were collected. To clearly distinguish these defects from basic defects, specific defect data refers to defects that occur under specific geological conditions or complex physical and chemical alternating effects. Unlike basic defects such as cracks and leaks, which manifest as single fractures or water stains, specific defects not only form unique textures, shapes, and colors on the surface, but are often accompanied by deep material deterioration, such as localized concrete carbonization and steel reinforcement corrosion and expansion. First, feature recognition was performed on these specific defect data. The texture characteristics of concrete carbonization were rough surface, black color, patchy distribution, and dark gray color. The texture characteristics of steel reinforcement corrosion and expansion were peeling surface, reddish color, dotted or linear shape, and reddish-brown color. Combining these texture, shape, and color characteristics, a data synthesis blueprint was generated. Then, based on this data synthesis blueprint, a conditional generative adversarial network was used to generate diverse synthetic defect data, such as images simulating different degrees of carbonization and different corrosion ranges. The generated synthetic disease data and the original specific disease data are stored in the same set and shuffled to create a model training package. Finally, an initial specific disease database model is constructed, and the model is trained using the training package until the model can accurately identify all specific disease data in the training package, resulting in the trained specific disease database model.

[0024] Based on the engineering design parameters within the target tunnel, the internal structure data of the tunnel is obtained. The specific steps for generating the UAV's identification route based on this internal structure data are as follows: Based on the engineering design parameters within the target tunnel, parameter extraction is performed on the engineering design parameters to obtain tunnel structure data and appearance construction data; Based on tunnel construction data and exterior construction data, the internal shape and size parameters of the tunnel are obtained, and the internal structure data is obtained based on the shape and size parameters. Obtain the single inspection range and fuselage size parameters of the drone, and combine the internal structure data, single inspection range and fuselage size parameters to generate the initial working route of the drone; The frequency of vehicle traffic within the tunnel is obtained, and the initial working route is adjusted based on the frequency of vehicle traffic to generate the drone's recognition working route.

[0025] In practice, taking a traffic tunnel connecting City A and City B as an example, based on the engineering design parameters of the target tunnel, such as construction drawings and structural design specifications, tunnel structural data and external construction data are extracted. Tunnel structural data includes the arch curvature, sidewall height, and road width, while external construction data includes concrete strength, waterproofing material, and surface treatment methods. Based on this data, the internal shape and dimensional parameters of the tunnel are obtained, such as an arch radius of 5 meters, a sidewall height of 3 meters, and a road width of 10 meters, thus obtaining the tunnel's internal structural data. Next, the single inspection range of the drone is determined to be 10 meters in diameter, and the drone's dimensions are 1 meter long, 0.5 meters wide, and 0.3 meters high. Combining the tunnel's internal structural data, the drone's single inspection range, and the drone's dimensions, an initial working route for the drone is generated, covering every inspection point along the entire tunnel length. Finally, the frequency of vehicle traffic within the tunnel is obtained, such as 50 vehicles per hour. The initial working route is adjusted based on this frequency to avoid peak traffic periods, generating the final drone recognition working route.

[0026] The drone conducts inspections according to the identified work route, simultaneously acquiring visible light and infrared thermal images of the target tunnel's exterior, and identifying defect points. The specific steps are as follows: The drone conducts inspections according to the identified work route, and collects images and thermal imaging data of the inner wall of the target tunnel, respectively, to obtain visible light images and infrared thermal images. Based on visible light images, the brightness and contrast of different regions in the visible light images are identified to obtain underexposed and overexposed areas. Local enhancement is applied to underexposed areas, and local suppression is applied to overexposed areas to obtain a visible light image of the target. Image recognition is performed on the visible light image of the target to identify abnormal image areas in the image; Based on infrared thermal imaging images, areas with temperature differences are identified to obtain abnormal temperature areas; By combining abnormal image areas and abnormal temperature areas, the disease points are obtained.

[0027] In application, taking a traffic tunnel connecting cities A and B as an example, the drone inspects the tunnel according to the generated identification route. During the inspection, the drone acquires both visual and thermal imaging data of the tunnel's inner wall, obtaining visible light and infrared thermal images respectively. Based on the visible light image, the drone identifies the brightness and contrast of different areas, marking some areas as too dark and others as too bright, respectively. Local enhancement is applied to the dark areas to increase brightness, while local suppression is applied to the overexposed areas to reduce brightness, resulting in a processed target visible light image. Image recognition is then performed on the target visible light image, identifying areas with color or texture anomalies, marking them as abnormal image areas. Based on the infrared thermal image, areas with temperature differences are identified, such as areas with temperatures significantly higher or lower than the surrounding areas, marking them as abnormal temperature areas. Finally, by combining the abnormal image and abnormal temperature areas, the location of defects, such as cracks and leaks, is determined.

[0028] The steps for performing image recognition on a target visible light image to identify abnormal image regions are as follows: Obtain dust images, water stain images, and old trace images inside the tunnel, and extract dust features from the dust images, water stain features from the water stain images, and old trace features from the old trace images. By combining dust characteristics, water stain characteristics, and old trace characteristics, environmental disturbance factor characteristics are generated. Based on the characteristics of environmental interference factors, interference factors are filtered from visible light images to obtain clean target visible light images. Then, image recognition is performed on the processed target visible light images to identify abnormal image areas in the images.

[0029] In application, taking a traffic tunnel connecting City A and City B as an example, dust, water stains, and signs of wear were collected inside the tunnel. The dust samples are characterized by granular particles and a grayish-white color; the water stain samples are characterized by flakes and a dark gray color; and the signs of wear are characterized by patches and a dark yellow color. These dust, water stain, and signs of wear features are combined to generate environmental interference factor features. Based on these features, the collected visible light images are filtered to remove interference from dust, water stains, and signs of wear, resulting in a clean target visible light image. Image recognition is then performed on the processed target visible light image, identifying areas with cracks, peeling, and other abnormalities, which are marked as abnormal areas. For example, a crack approximately 2 meters long and 0.5 centimeters wide was identified on a section of the tunnel sidewall, with obvious color changes and texture anomalies around the crack.

[0030] After simultaneously acquiring visible light and infrared thermal images of the target tunnel's interior and exterior, and identifying the defects, the process also includes: Based on the disease points, identify the actual size and type of disease, and conduct a risk assessment based on the actual size and type of disease to obtain an initial risk value; Determine whether the initial risk value is less than the preset standard risk value. If the initial risk value is determined to be less than the standard risk value, then track and monitor the disease points to obtain disease change data over a period of time. Based on the disease change data, the disease evolution trend of the disease points is obtained, and the risk change level label is obtained based on the disease evolution trend. The risk change level label is then affixed to the location of the disease point.

[0031] In application, taking a traffic tunnel connecting City A and City B as an example, based on identified defects, such as a crack, the actual size of the defect is identified as 2 meters in length and 0.5 centimeters in width, and the defect type is identified as a structural crack. A risk assessment is performed based on the actual size and defect type, resulting in an initial risk value of medium. It is then determined whether the initial risk value is lower than a preset standard risk value, which is set as low risk. Since the initial risk value is medium and higher than the standard risk value, the defect point is monitored. Over a period of time, such as one month, data on the defect point is collected regularly to obtain defect change data, such as the crack length increasing to 2.5 meters and the width increasing to 0.8 centimeters. Based on the defect change data, the evolution trend of the defect is analyzed, revealing a slow expansion trend. Based on this trend, a post-calculation risk change label is obtained, labeled "Risk Increase," and this label is affixed to the location of the defect to remind maintenance personnel to pay close attention.

[0032] Based on the basic disease database model and the special disease database model, the steps for identifying disease points and performing correlation analysis to obtain the risk of compound diseases are as follows: Based on the basic disease database model and the special disease database model, disease points are identified to obtain the current disease type and current disease range of the disease points; Based on the current disease type and the current disease range, data are correlated in the basic disease database model and the special disease data model to obtain the target correlated data; Extract disease data from the target associated data to obtain associated disease data, and match the associated disease data according to the current disease type to obtain multiple matching values; Determine whether the matched value exceeds the preset standard value. If the matched value exceeds the standard value, mark the associated disease data as candidate associated disease data. The explicit features of the candidate related disease data are extracted, and the explicit features are compared and screened with visible light images and infrared thermal imaging images to obtain the target related disease data; By combining the current disease type, the current disease range, and the target related disease data, composite disease parameters are obtained. A risk assessment is then conducted on these composite disease parameters to obtain the composite disease risk.

[0033] In application, taking a traffic tunnel connecting City A and City B as an example, based on the basic defect database model and the special defect database model, the identified defect points are identified, and the current defect type is determined to be structural cracks, with a current defect range of 2.5 meters in length and 0.8 centimeters in width. Based on the current defect type and range, data association is performed in the basic defect database model and the special defect database model to obtain target associated data, such as the potential for cracks to cause leakage, concrete spalling, and other related defects. Defect extraction is performed on the target associated data to obtain associated defect data, such as specific information on leakage and spalling. The associated defect data is matched according to the current defect type, resulting in multiple matching values; for example, the matching value for cracks and leakage is 0.8, and the matching value for cracks and spalling is 0.6. It is then determined whether the matching value exceeds a preset standard value of 0.7; if it does, the associated defect data is marked as candidate associated defect data. The explicit features of candidate associated disease data, such as the color and temperature characteristics of the leakage area, are extracted and compared with visible light images and infrared thermal imaging images to obtain the target associated disease data, such as the leakage area. Combining the current disease type, current disease extent, and target associated disease data, composite disease parameters are obtained. A risk assessment is then performed on these composite disease parameters, resulting in a high-risk composite disease.

[0034] The steps for constructing a 3D point cloud model of the target tunnel, combining the 3D point cloud model with the risk of complex defects, constructing a 3D defect map, and formulating a maintenance plan based on the 3D defect map are as follows: Based on internal structural data and infrared thermal imaging images, the external structure of the tunnel's internal surface is scanned to generate point cloud data of the tunnel. Based on point cloud data, the interior of the target tunnel is modeled to generate a 3D point cloud model of the target tunnel. The composite disease parameters are reproduced in a three-dimensional point cloud model to obtain a three-dimensional point cloud model of the target disease. By combining the 3D point cloud model of the target disease and the risk of complex diseases, a 3D disease map is constructed. The causes of the diseases are determined based on the 3D disease map, and a maintenance plan is formulated based on the causes of the diseases to obtain a maintenance treatment plan.

[0035] In application, taking a traffic tunnel connecting City A and City B as an example, based on the tunnel's internal structural data and infrared thermal imaging images, the external structure of the tunnel's internal surface is scanned to generate point cloud data. The point cloud data includes the three-dimensional coordinates of the tunnel's inner wall and surface temperature information. Based on the point cloud data, the interior of the target tunnel is modeled, generating a three-dimensional point cloud model that accurately displays the tunnel's geometry and temperature distribution. Complex defect parameters are reproduced in the three-dimensional point cloud model, such as marking the location and extent of defects like cracks and leaks, resulting in a three-dimensional point cloud model of the target defects. Combining the three-dimensional point cloud model of the target defects and the risks associated with the complex defects, a three-dimensional defect map is constructed, visually displaying the distribution and risk level of the defects. The causes of the defects are determined based on the three-dimensional defect map; for example, cracks may be due to structural stress concentration, and leaks may be due to aging of the waterproofing layer. Based on the causes of the defects, repair and treatment plans are developed, such as grouting to reinforce cracks and repairing the waterproofing layer in leaking areas to ensure the safety of the tunnel structure.

[0036] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for detecting tunnel surface defects based on artificial intelligence, characterized in that, include: Acquire common defect data, tunnel characteristics, and engineering design parameters of the target tunnel from different tunnels, construct a defect database model, and perform transfer learning on the defect database model based on the engineering design parameters to obtain a basic defect database model. Acquire special disease data from different tunnels, identify the texture, shape and color features of the special disease data, and generate diverse synthetic disease data based on conditional generative adversarial network. Combine the special disease data and the synthetic disease data to train the model and obtain a special disease database model. Based on the engineering design parameters within the target tunnel, the internal structure data of the tunnel is obtained, and the identification working route of the UAV is generated according to the internal structure data. The drone conducts inspections according to the identified work route, and simultaneously collects visible light images and infrared thermal images of the target tunnel's exterior, and identifies the defects. Based on the basic disease database model and the special disease database model, the disease points are identified and correlation analysis is performed to obtain the risk of compound diseases. Based on the internal structure data and the infrared thermal imaging image, a three-dimensional point cloud model of the target tunnel is constructed. Combining the three-dimensional point cloud model and the composite disease risk, a three-dimensional disease map is constructed. Based on the three-dimensional disease map, a maintenance plan is formulated to obtain a maintenance treatment plan.

2. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 1, characterized in that, The steps for constructing a disease database model and performing transfer learning on the disease database model based on the engineering design parameters to obtain a basic disease database model are as follows: Based on the tunnel characteristics, the service life, design purpose, usage scenarios, and appearance of different tunnels are obtained. Based on the service life, design purpose, usage scenario, and appearance shape, a database model framework for multi-source data fusion is generated; Based on the general disease data, disease identification is performed on the general disease data to obtain multiple general disease types, as well as the disease severity value and disease treatment plan corresponding to each general disease type; Based on the general disease type, the disease severity value, and the disease treatment plan, generate the content of a database model that integrates multi-source data, and construct a disease database model by combining the database model framework and the database model content. Based on the engineering design parameters of the target tunnel, the disease database model is migrated, and the migrated data is re-integrated to generate a basic disease database model.

3. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 2, characterized in that, The steps of identifying the texture, shape, and color features of the specific disease data, generating diverse synthetic disease data based on a conditional generative adversarial network, and training a model by combining the specific disease data and the synthetic disease data to obtain a specific disease database model are as follows: Feature recognition is performed on the special disease data to obtain the texture features, shape features and color features of the special disease data respectively, and the texture features, shape features and color features are combined to generate a data synthesis blueprint; Based on the data synthesis blueprint, an adversarial network is generated according to the conditions to produce diverse synthetic disease data; The synthetic disease data and the special disease data are stored in the same set and shuffled to generate a model training package; An initial special disease database model is constructed, and the initial special disease database model is trained using the model training package until the initial special disease database model can identify all special disease data in the model training package, thus obtaining the special disease database model.

4. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 3, characterized in that, The steps for obtaining the internal structure data of the tunnel based on the engineering design parameters within the target tunnel, and generating the UAV's identification route based on the internal structure data, are as follows: Based on the engineering design parameters within the target tunnel, parameter extraction is performed on the engineering design parameters to obtain tunnel structure data and external construction data. Based on the tunnel construction data and the exterior construction data, the internal shape and size parameters of the tunnel are obtained, and the internal structure data is obtained based on the shape and size parameters. The drone's single inspection range and fuselage size parameters are obtained. Combined with the internal structure data, the single inspection range, and the fuselage size parameters, the drone's initial working route is generated. The frequency of vehicle traffic within the tunnel is obtained, and the initial working route is adjusted based on the vehicle traffic frequency to generate an identification working route for the drone.

5. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 4, characterized in that, The drone conducts inspections according to the identified work route, simultaneously acquiring visible light and infrared thermal images of the target tunnel's exterior, and identifying defect points. The specific steps are as follows: The drone conducts inspections according to the identified working route, and collects images and thermal imaging data of the inner wall of the target tunnel, respectively, to obtain visible light images and infrared thermal imaging images. Based on the visible light image, the brightness and contrast of different regions in the visible light image are identified to obtain underexposed and overexposed regions. The underexposed areas are locally enhanced, and the overexposed areas are locally suppressed to obtain a visible light image of the target. Image recognition is performed on the target visible light image to identify abnormal image regions present in the image; Based on the infrared thermal imaging image, areas with temperature differences are identified to obtain abnormal temperature areas; By combining the abnormal image area and the abnormal temperature area, the disease points are obtained.

6. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 5, characterized in that, The steps of performing image recognition on the target visible light image to obtain abnormal image regions in the image are as follows: Obtain dust image samples, water stain image samples, and old trace image samples inside the tunnel, and extract the dust features of the dust image samples, the water stain features of the water stain image samples, and the old trace features of the old trace image samples, respectively. By combining the dust characteristics, the water stain characteristics, and the old trace characteristics, environmental interference factor characteristics are generated; The visible light image is filtered for interference factors based on the environmental interference factor characteristics to obtain a clean target visible light image. The processed target visible light image is then used for image recognition to identify abnormal image areas.

7. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 6, characterized in that, After simultaneously acquiring visible light and infrared thermal images of the target tunnel's interior and exterior, and identifying the defects, the process also includes: Based on the disease points, the actual size and disease type of the disease points are identified, and a risk assessment is performed based on the actual size and disease type to obtain an initial risk value; Determine whether the initial risk value is less than the preset standard risk value. If the initial risk value is less than the standard risk value, then track and monitor the disease point to obtain disease change data over a period of time. Based on the disease change data, the disease evolution trend of the disease point is obtained, and the risk change degree label is obtained by extrapolation based on the disease evolution trend. The risk change degree label is then affixed to the location of the disease point.

8. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 7, characterized in that, Based on the aforementioned basic disease database model and the aforementioned special disease database model, the steps for identifying disease points and performing correlation analysis to obtain the risk of compound diseases are as follows: Based on the basic disease database model and the special disease database model, the disease points are identified to obtain the current disease type and current disease range of the disease points; Based on the current disease type and the current disease range, data association is performed in the basic disease database model and the special disease data model to obtain target associated data; Disease extraction is performed on the target associated data to obtain associated disease data, and the associated disease data is matched according to the current disease type to obtain multiple matching values; Determine whether the matching value exceeds a preset standard value. If the matching value exceeds the standard value, mark the associated disease data as candidate associated disease data. The explicit features of the candidate associated disease data are extracted, and the explicit features are compared and screened with the visible light image and the infrared thermal image to obtain the target associated disease data; By combining the current disease type, the current disease range, and the target associated disease data, composite disease parameters are obtained. A risk assessment is then performed on the composite disease parameters to obtain the composite disease risk.

9. The method for detecting tunnel appearance defects based on artificial intelligence according to claim 8, characterized in that, The steps for constructing a 3D point cloud model of the target tunnel, combining the 3D point cloud model with the composite defect risk, constructing a 3D defect map, and formulating a maintenance plan based on the 3D defect map are as follows: Based on the internal structure data and the infrared thermal imaging image, the external structure of the tunnel's internal surface is scanned to generate point cloud data of the tunnel. Based on the point cloud data, the interior of the target tunnel is modeled to generate a three-dimensional point cloud model of the target tunnel. The composite disease parameters are reproduced in the three-dimensional point cloud model to obtain the three-dimensional point cloud model of the target disease. By combining the target disease 3D point cloud model and the composite disease risk, a 3D disease map is constructed, and the cause of the disease is determined based on the 3D disease map. Based on the cause of the disease, a maintenance plan is formulated to obtain a maintenance treatment plan.

Images