Building exterior facing apparent damage survey method based on unmanned aerial vehicle

Abstract

The invention discloses a building exterior facing apparent damage survey method based on an unmanned aerial vehicle, and the method comprises the steps: enabling a trained damage detection instance segmentation deep learning model and a trained material classification neural network model to be deployed in an unmanned aerial vehicle, and carrying out the deployment of the unmanned aerial vehicle; the unmanned aerial vehicle is used for routing inspection route planning, close routing inspection collection, damage detection and segmentation, pixel size calculation, facing material identification, spatial parameter calculation and damage evaluation treatment, so that the apparent damage of the exterior facing of the building is intelligently detected and the damage degree and grade are evaluated through the two dimensions of damage and material, and a treatment suggestion is given. The method realizes closed-loop management of detection-evaluation-disposal, and has the advantages of high detection efficiency and high precision.

Description

Technical Field

[0001] The present invention relates to the field of historical building maintenance, and particularly to a method for surveying the apparent damage of the exterior finish of a building based on an unmanned aerial vehicle (UAV). Background Art

[0002] The materials, techniques, and styles of the exterior finishes of historical buildings are key protection contents. Their deterioration (such as brick powdering, component rusting, and finish layer peeling) will directly lead to the loss of cultural relic value. Therefore, it is necessary to survey the apparent damage of the exterior finishes of historical buildings to formulate maintenance plans. Currently, the apparent damage detection of the exterior finishes of historical buildings can be carried out through UAV inspections. Its disadvantages are as follows: The recognition model based on the object detection algorithm cannot complete the quantitative evaluation of the deterioration degree of the finishes. The quantitative evaluation still relies on the experience judgment of professionals, and there are certain deviations in the evaluation results of different institutions, which are difficult to support scientific decision-making. In addition, UAV inspections only support the single-time identification of cracks and peeling damages and do not support the data management and analysis of multiple identification results. The current UAV inspection scheme lacks the closed-loop management of "detection - evaluation - disposal". The classification of the apparent damage levels of the exterior finishes still relies on manual secondary determination, and the disposal suggestions have not formed a knowledge base, which is difficult to provide guidance for the maintenance management decision-making and digital management of historical buildings. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for surveying the apparent damage of the exterior finish of a building based on an unmanned aerial vehicle, so as to solve the problems that when the UAV inspection identifies the apparent damage, it cannot conduct a quantitative evaluation, and professional personnel need to make a secondary judgment based on experience, resulting in a large subjective judgment error and being difficult to provide support for scientific decision-making, and to solve the problem that the current UAV inspection lacks the closed-loop management of "detection - evaluation - disposal".

[0004] To solve the above technical problems, the present invention provides a method for surveying the apparent damage of the exterior finish of a building based on an unmanned aerial vehicle, including: UAV deployment, deploying the trained deep learning model for instance segmentation of damage detection and the neural network model for material classification in the UAV; Inspection route planning, collecting aerial photos of the building roof and exterior facade by the UAV using the oblique photogrammetry method, resolving and generating a rough building model through the aerial triangulation method, and generating an inspection route plan for the rough building model through the path planning algorithm based on heuristic search; Proximity inspection collection, collecting all-image data of the exterior finish of the building by flying the UAV to each flight point to take pictures of the building; Damage detection and segmentation, the UAV automatically cuts the collected images of the exterior finish of the building through the trained deep learning model for instance segmentation of damage detection and automatically identifies the apparent damage situation of the cut images, and forms a damage mask by automatically marking the apparent damage area Pixel size calculation. The drone calculates the pixel size of the damaged area by calculating the masked pixel points of the damage mask in the image data segmented by computer vision methods. Finishing material identification. The drone identifies the material type of the apparent damage area by using a material classification neural network model to process the image data of the apparent damage area. Spatial parameter solution. The drone calculates the exterior orientation elements of the image with the apparent damage area and the ground coordinates of the encrypted points through aerial triangulation, and combines the interior orientation elements of the drone's camera to obtain the actual spatial coordinate positions of the damaged points in the apparent damage area and the pixel resolution of each image. Damage assessment and disposal. The drone converts the pixel size of the damaged area and the pixel resolution of the corresponding image to obtain the actual size of the apparent damage area. Based on the area size of the actual size of the apparent damage area and combined with the damage assessment theory of different finishing material types, the apparent damage area is evaluated according to a three-level evaluation system of minor, moderate, and severe. For minor damage, the disposal suggestion is to increase the inspection frequency for observation; for moderate damage, the disposal suggestion is to intervene with environmental measures to delay; for severe damage, the disposal suggestion is to immediately repair it with a repair plan.

[0005] Furthermore, the method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention further includes, after damage detection and assessment: Damage confidence filtering. The drone screens and filters the damage data of the same damaged area in different damaged images. For damages with the deviation of the damage center position within a predetermined deviation range and the same damage type, they are classified into the same damage data set. Then, all the data in the data set are screened, and the one with the damage position center closer to the shooting center is selected as the trusted image, and the damage detection result of the trusted image is used as the detection and assessment result of the apparent damage.

[0006] Furthermore, the method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention further includes: Three-dimensional model construction. After the drone obtains the exterior orientation elements of each image, it performs model orientation, unifies the coordinate systems of multiple images, densely matches adjacent images based on features to obtain homologous points, thereby generating three-dimensional point cloud data of the building exterior finish. A triangular mesh is constructed through the three-dimensional point cloud data, and the image texture information is mapped onto the surface of the triangular mesh to establish a three-dimensional model of the building exterior finish. Furthermore, the method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention further includes: Damage model management: The drone constructs guiding points on the 3D model of the building exterior finish according to the actual spatial coordinates of the damage points. After clicking on the guiding points, information such as finish material, damage type, damage parameters, damage grade, and disposal suggestions can be displayed. Furthermore, the method for surveying the apparent damage of the building exterior finish based on the drone provided by the present invention further includes: Damage data update: Update the survey results of multiple damage data of the same building to the 3D model of the building exterior finish.

[0007] Furthermore, for the method for surveying the apparent damage of the building exterior finish based on the drone provided by the present invention, the damage data update method includes: The three-dimensional coordinate errors of multiple data acquisitions can be eliminated by setting permanent image control points around the building and incorporating them into the aerial triangulation solution. After obtaining new damage data, it is determined whether it belongs to the front and back inspection results of the same damage by judging whether the deviation from the damage center position of the previous inspection is within a predetermined deviation range and the damage type is the same. For new damage, a new piece of data is created, and for old damage, the damage data is updated, and the original data is included in the historical evolution of this piece of damage, completing the update management of the damage data on the 3D model of the building exterior finish.

[0008] Furthermore, for the method for surveying the apparent damage of the building exterior finish based on the drone provided by the present invention, in the calculation of pixel size, For area-type damage, the pixel size of the damage area is obtained by calculating the number of pixel points within the area. For crack-type damage, the pixel size of the damage area is obtained by calculating the maximum value of the cumulative pixel points in the crack length direction and the number of pixel points in the width direction through the centerline method.

[0009] Furthermore, for the method for surveying the apparent damage of the building exterior finish based on the drone provided by the present invention, in the damage detection segmentation, the images collected by the drone are cut into detection units of 1024×1024 for parallel calculation, and then adjacent similar damages are merged and the local coordinate system is transformed into the global coordinate system to realize the automatic segmentation of the drone images and the automatic detection of damages.

[0010] Furthermore, for the method for surveying the apparent damage of the building exterior finish based on the drone provided by the present invention, in the drone deployment, the damage detection instance segmentation deep learning model is trained by collecting and annotating one or more of the damage images including cracking, defect, epiphyte, rust spot, efflorescence, weathering, decay, peeling as a data set, so that the trained damage detection instance segmentation deep learning model can automatically identify the damage type according to the input image and perform the function of instance extraction.

[0011] Furthermore, in the method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention, during the deployment of the drone, the material classification neural network model is trained for classification tasks by collecting and annotating one or more of the damage images containing stone, Taishan bricks, fair - faced bricks, real stone paint, wood veneer, washed stone, terrazzo, and paint as a data set, so that the trained material classification neural network model has the function of automatically identifying the type of finish material according to the input image.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: The method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention deploys a trained deep - learning model for damage detection instance segmentation and a material classification neural network model in the drone. The drone conducts close - range inspections of the building along the planned inspection route to collect image data of the exterior finish. The collected images of the exterior finish are automatically segmented into small - sized images by the trained deep - learning model for damage detection instance segmentation, and the damage conditions of the segmented small - sized images are automatically detected. The pixel size of the damage area is calculated through computer vision technology; the material type of the damage area is automatically identified by the trained material classification neural network model; the spatial coordinate position of the damage area and the pixel resolution of each image are calculated through spatial parameter resolution. The actual size of the apparent damage area is determined by the pixel size and pixel resolution of the damage area, and the apparent damage area is evaluated in combination with the material type and disposal suggestions are given, so as to realize the closed - loop management of detection - evaluation - disposal of the damage of the building exterior finish through the drone.

[0013] The method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention uses a two - stage deep - learning model for damage recognition - material recognition in the drone. While reducing the algorithm complexity, it realizes the accurate quantitative evaluation of losses, supports damage grading and disposal suggestions. Through the digital evaluation of the number of pixel points in the damage area and the evaluation in combination with the material type, it avoids the problem that manual secondary judgment based on experience for the images collected by the drone has large subjective judgment errors and is difficult to provide support for scientific decision - making.

[0014] The method for surveying the apparent damage of the building exterior finish based on a drone provided by the present invention conducts all surveys of the apparent damage of the building exterior finish in the drone, without the need to build a scaffolding or high - altitude operation platform, reducing the safety risks and efficiency bottlenecks of traditional manual high - altitude inspections, and realizing non - contact comprehensive inspections, with the advantages of high efficiency and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flowchart of the method for surveying the apparent damage of the building exterior finish based on a drone. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The present invention will be described in detail below with reference to the accompanying drawings: The advantages and features of the present invention will be more clearly understood according to the following description.

[0017] Please refer to Figure 1 , an embodiment of the present invention provides a method for surveying the apparent damage of building exterior finishes based on an unmanned aerial vehicle, which at least includes the steps of unmanned aerial vehicle deployment, inspection route planning, close inspection and acquisition, damage detection and segmentation, pixel size calculation, finish material identification, spatial parameter calculation, and damage assessment and disposal, wherein: Unmanned aerial vehicle deployment: Deploy the trained deep learning model for damage detection instance segmentation and the material classification neural network model in the unmanned aerial vehicle. Among them, the deep learning model for damage detection instance segmentation is trained by collecting and annotating one or more of the damage images including cracking, defect, epiphyte, rust spot, efflorescence, weathering, decay, peeling as a data set, so that the trained deep learning model for damage detection instance segmentation can automatically identify the damage type and perform instance extraction according to the input image. Among them, the material classification neural network model is trained for classification tasks by collecting and annotating one or more of the damage images including stone, Taishan brick, fair-faced brick, real stone paint, wood finish, washed stone, terrazzo, paint as a data set, so that the trained material classification neural network model has the function of automatically identifying the type of finish material according to the input image.

[0018] Inspection route planning: Use the unmanned aerial vehicle to collect aerial photos of the building roof and exterior facade by means of oblique photogrammetry, calculate and generate a rough building model by means of aerial triangulation, and generate an inspection route plan for the rough building model through a path planning algorithm of heuristic search. Among them, the path planning algorithm of heuristic search uses the A* algorithm, combines the curvature analysis of the building exterior facade, and generates a "zigzag" close inspection route (3-5m), which can completely cover the building exterior facade and meet the requirements of the detection safety distance and acquisition accuracy of the unmanned aerial vehicle.

[0019] Close inspection and acquisition: Collect all the image data of the building exterior finish by flying the unmanned aerial vehicle to each flight point to take pictures of the building. Among them, when the unmanned aerial vehicle is flying, it uses the onboard binocular vision sensor, infrared TOF and ultrasonic sensors to capture the distance, contour and movement trajectory of obstacles in real time to achieve precise obstacle avoidance. Among them, the ToF (Time-of-Flight) sensor is a sensor that calculates the distance based on the time of flight, and it calculates the distance based on the time difference between the signal emission and its return to the sensor after being reflected by the object.

[0020] Damage detection and segmentation: The drone automatically cuts the images of the building exterior finish collected by the trained instance segmentation deep learning model for damage detection, and automatically identifies the apparent damage conditions of the cut images. The damage mask is formed by automatically marking the apparent damage area. In damage detection and segmentation, since the images taken by the drone have a high resolution and are taken at a certain distance from the wall, the amount of finish damage information in them may be very small, showing the characteristics of small targets. Directly inputting them into the deep learning model results in low detection accuracy. Therefore, to improve the detection accuracy, the images collected by the drone can be cut into detection units of 1024×1024 for parallel computing, and then adjacent similar damages are merged and the local coordinate system is transformed into the global coordinate system to achieve automatic segmentation of drone images and automatic damage detection.

[0021] Pixel size calculation: The drone calculates the pixel size of the damage area by computing the mask pixel points of the damage mask in the image data segmented by computer vision methods. To accurately calculate the pixel size for different damage contours, in pixel size calculation, for area-type damages, the pixel size of the damage area is obtained by calculating the number of pixel points within the area; for crack-type damages, the maximum value of the cumulative pixel points in the crack length direction and the number of pixel points in the width direction is obtained by the center line method to get the pixel size of the damage area. By using different methods to calculate different damage contours, the accuracy of calculating the pixel size of each damage area can be improved.

[0022] Finish material identification: The drone identifies the material type of the apparent damage area by using a material classification neural network model for the image data of the apparent damage area.

[0023] Spatial parameter solution: The drone calculates the exterior orientation elements of the image with the apparent damage area and the ground coordinates of the encrypted points through aerial triangulation methods, and combines the interior orientation elements of the drone's camera to obtain the actual spatial coordinate positions of the damage points in the apparent damage area and the pixel resolution of each image.

[0024] Damage assessment and disposal: The drone calculates the actual size of the apparent damage area by converting the pixel size of the damage area obtained and the pixel resolution of the corresponding image, and evaluates the apparent damage area according to the area size of the actual size of the apparent damage area and in combination with the damage assessment theory of different finishing material types, using a three-level evaluation system of minor, moderate, and severe; for minor damage, the disposal suggestion of increasing the inspection frequency for observation is adopted, for moderate damage, the disposal suggestion of intervening and delaying through environmental measures is adopted; for severe damage, the disposal suggestion of immediate repair is adopted. Since the theory of apparent damage assessment and disposal is not only related to the type of damage but also closely related to the nature of the finishing material, it is also necessary to extract the type of finishing material. The aerial image may contain different finishing materials. By extracting the masked area obtained in the damage detection segmentation and inputting it into the finishing material recognition model, the type of the finishing material at the damage location is obtained.

[0025] The method for inspecting the apparent damage of the building exterior finish based on a drone provided by the embodiment of the present invention deploys a trained deep learning model for damage detection instance segmentation and a material classification neural network model in the drone. The drone closely inspects and collects image data of the exterior finish according to the planned inspection route of the building. Each collected image of the exterior finish is automatically segmented into small-sized images by the trained deep learning model for damage detection instance segmentation, and the damage situation of the segmented small-sized images is automatically detected. The pixel size of the damage area is calculated through computer vision technology; the material type of the damage area is automatically identified by the trained material classification neural network model; the spatial coordinate position of the damage area and the pixel resolution of each image are obtained through spatial parameter calculation. The actual size of the apparent damage area is determined by the pixel size and pixel resolution of the damage area, and the apparent damage area is evaluated in combination with the material type and disposal suggestions are given, so as to realize the closed-loop management of detection-evaluation-disposal of the damage situation of the building exterior finish through the drone.

[0026] The method for inspecting the apparent damage of the building exterior finish based on a drone provided by the embodiment of the present invention deploys the trained deep learning model for damage detection instance segmentation and the material classification neural network model in the drone for drone deployment. Through the drone, inspection route planning, close inspection and collection, damage detection and segmentation, pixel size calculation, finishing material identification, spatial parameter calculation, damage assessment and disposal are carried out, so as to intelligently detect the apparent damage of the building exterior finish in two dimensions of damage and material, evaluate the damage degree level and give treatment suggestions, realizing the closed-loop management of detection-evaluation-disposal, and having the advantages of high detection efficiency and high precision.

[0027] The method for surveying the apparent damage of the building exterior finish based on an unmanned aerial vehicle (UAV) provided by the embodiment of the present invention uses a two-stage deep learning model for damage recognition - material recognition in the UAV, which can reduce the algorithm complexity while achieving accurate quantitative evaluation of losses, supporting damage grading and disposal suggestions. Through the digital evaluation of the number of pixel points in the damaged area and the evaluation in combination with the material type, it avoids the problem that manual secondary judgment based on experience for the images collected by the UAV has a large subjective judgment error and is difficult to provide support for scientific decision-making.

[0028] The method for surveying the apparent damage of the building exterior finish based on an unmanned aerial vehicle (UAV) provided by the embodiment of the present invention conducts all the surveys of the apparent damage of the building exterior finish in the UAV, without the need to erect a scaffolding or an aerial work platform, reducing the safety risks and efficiency bottlenecks of traditional manual high-altitude detection, achieving non-contact comprehensive detection, and having the advantages of high efficiency and intelligence.

[0029] Please refer to Figure 1 , in order to improve the accuracy of the survey and detection results, the method for surveying the apparent damage of the building exterior finish based on an unmanned aerial vehicle (UAV) provided by the embodiment of the present invention may further include, after the damage detection and evaluation: Damage confidence filtering. The UAV screens and filters the damage data of the same damaged area in different damaged images. For the damages with the deviation of the damage center position within the predetermined deviation range and the same damage type, they are classified into the same damage data set, and then all the data in the data set are screened. The image with the damage position center closer to the shooting center is selected as the trusted image, and the damage detection result of the trusted image is used as the detection and evaluation result of the apparent damage.

[0030] In order to conduct visual management of the apparent damage situation of the building exterior finish in the whole cycle, the method for surveying the apparent damage of the building exterior finish based on an unmanned aerial vehicle (UAV) provided by the embodiment of the present invention may further include: Three-dimensional model construction. After the UAV obtains the exterior orientation elements of each image, it conducts model orientation, unifies the coordinate systems of multiple images, conducts dense matching based on features for adjacent images, obtains homologous points to generate the three-dimensional point cloud data of the building exterior finish, constructs a triangular mesh through the three-dimensional point cloud data, and maps the image texture information to the surface of the triangular mesh to establish a three-dimensional model of the building exterior finish. Specifically, a triangular mesh is constructed through the discrete points in the three-dimensional point cloud data, and the image texture information is mapped to the surface of the triangular mesh to form a three-dimensional model of the building exterior finish. Among them, the three-dimensional model construction can be carried out in the UAV or on the PC side.

[0031] Damage model management: The drone constructs guiding points on the 3D model of the building exterior finish according to the actual spatial coordinates of the damage points. After clicking on the guiding points, information such as the finish material, damage type, damage parameters, damage level, and disposal suggestions can be displayed. By loading the above-displayable damage situation and disposal suggestion information onto the 3D model of the building exterior finish, a 3D digital twin model of the building exterior finish is formed, enabling visual management of the building exterior finish and achieving the goal of efficient damage management based on the twin model.

[0032] To manage and analyze the evolution process of the inspection results of the building exterior facade multiple times, the method for inspecting the apparent damage of the building exterior finish based on a drone provided by the embodiments of the present invention may further include: Damage data update: Update the inspection results of the damage data of the same building onto the 3D model of the building exterior finish. Specifically, it may include: The 3D coordinate errors of multiple data collections can be eliminated by setting permanent image control points around the building and incorporating them into the aerial triangulation solution. After obtaining new damage data, it is determined whether it belongs to the front and back inspection results of the same damage by judging whether the deviation from the damage center position of the previous inspection is within a predetermined deviation range such as 3 cm and the damage type is the same. For new damage, a new piece of data is created, and for old damage, the damage data is updated, and the original data is included in the historical evolution of this piece of damage, completing the update management of the damage data on the 3D model of the building exterior finish. Thus, the problem that the drone only supports single recognition for the inspection of the building exterior finish and cannot manage and analyze the results of multiple recognitions is solved. Simply put, for multiple damage detections of the same building, the data can also be updated and managed according to the technical route of close inspection collection - damage detection segmentation - pixel size calculation - finish material identification - spatial parameter solution - damage assessment and disposal.

[0033] The method for inspecting the apparent damage of the building exterior finish based on a drone provided by the embodiments of the present invention, through the update of the damage data of the 3D digital twin model, integrates the spatio-temporal coordinates, material attributes, and damage evolution data of the damage situation of the building exterior finish, realizes the dynamic visual traceability of the damage location, degree, and development process, solves the problems of scattered traditional 2D drawing data and lack of relevance, and provides all-element and sustainable digital asset support for the preventive protection of historical building exterior finishes.

[0034] The main idea of the method for surveying the apparent damage of the building exterior finish based on drones provided by the embodiments of the present invention is to collect digital images of the building exterior finish based on the close-range photogrammetry technology of drones, analyze the damage types in the digital images and calculate the damage parameters through instance segmentation neural networks and computer vision methods, and assist the material classification neural network model to judge the material types of the segmented instance areas; through the aerial triangulation method, solve the spatial positions and parameters of the digital images, and construct the model base of the building; according to the spatial parameters of the digital images, the three-dimensional coordinates and actual damage sizes of the damage can be obtained, realizing the detection and evaluation, screening and filtering, twin management and dynamic update of the damage, and constructing an integrated solution of "dynamic monitoring - intelligent diagnosis - scientific decision-making", providing technical support for the scientific protection and value inheritance of the historical building exterior finish.

[0035] The present invention is not limited to the above specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention belong to the scope of protection of the present invention. Those skilled in the art can make other levels of modifications and changes to the present invention. Thus, if these modifications and changes of the present invention fall within the scope of the claims of the present invention, the present invention also intends to include these modifications and changes.

Claims

1. A method for inspecting the surface damage of building exterior surfaces based on drones, characterized in that: include: UAV deployment: deploy the trained damage detection instance segmentation deep learning model and material classification neural network model in the drone; Inspection route planning: using drones to collect aerial photos of building roofs and facades using oblique photogrammetry, using aerial triangulation to generate a rough model of the building, and using a heuristic search path planning algorithm to generate an inspection route plan for the rough model of the building; Close inspection and collection: drones fly to each waypoint to take pictures of the building and collect full image data of the building's exterior finishes; Damage detection and segmentation: The drone uses the trained damage detection instance segmentation deep learning model to automatically cut the collected images of the building exterior and automatically identify the apparent damage of the cut images, and automatically marks the apparent damage areas to form a damage mask; Pixel size calculation: the pixel size of the damaged area is obtained by calculating the mask pixels of the damage mask in the image data segmented by the drone using computer vision methods; Finishing material identification: The drone uses a material classification neural network model to identify the image data of the apparent damage area and obtain the material type of the apparent damage area; Spatial parameter solution: the drone calculates the exterior orientation elements of the image with the apparent damage area and the ground coordinates of the encrypted points through the aerial triangulation method, and combines the interior orientation elements of the drone's camera to obtain the actual spatial coordinate position of the damage point in the apparent damage area and the pixel resolution of each image; For damage assessment and disposal, the drone obtains the pixel size of the damaged area and the pixel resolution of the corresponding image to obtain the actual size of the apparent damaged area. Based on the actual size of the apparent damaged area and the damage assessment theory of different types of finishing materials, the apparent damaged area is evaluated according to a three-level evaluation system of slight, moderate and severe damage. For slight damage, the drone recommends increasing the inspection frequency for observation. For moderate damage, the drone recommends using environmental measures to delay the damage. For severe damage, the drone recommends using an immediate repair plan to repair the damage.

2. The method for inspecting the surface damage of building exterior surfaces based on drone according to claim 1 is characterized in that: After the damage detection assessment also includes: Damage confidence filtering: The drone screens and filters the damage data of the same damage area in different damage images. Damages with damage center position deviations within the predetermined deviation range and consistent damage types are classified into the same damage data set. All data in the data set are then screened, and the ones with damage position centers closer to the shooting center are selected as trust images. The damage detection results of the trust images are used as the detection and evaluation results of apparent damage.

3. The method for inspecting the apparent damage of exterior building surfaces based on an unmanned aerial vehicle according to claim 1 or 2, characterized in that: Also includes: To build a three-dimensional model, the drone obtains the exterior elements of each image and then orients the model. The coordinate systems of multiple images are unified, and adjacent images are densely matched based on features. Points with the same name are obtained to generate three-dimensional point cloud data of the building's exterior. A triangular mesh is constructed using the three-dimensional point cloud data, and the image texture information is mapped to the triangular mesh surface to establish a three-dimensional model of the building's exterior.

4. The method for inspecting the surface damage of building exterior surfaces based on drone according to claim 3 is characterized in that: Also includes: Damage model management: The drone constructs guide points on the three-dimensional model of the building's exterior finish based on the actual spatial coordinates of the damage point. Clicking the guide point will display the finish material, damage type, damage parameters, damage level, and disposal recommendations.

5. The method for inspecting the apparent damage of exterior building surfaces based on drones according to claim 4 is characterized in that: Also includes: Damage data update: update the multiple damage data survey results of the same building to the three-dimensional model of the building's exterior finishes.

6. The method for inspecting the surface damage of building exterior surfaces based on drones according to claim 5 is characterized in that: The damage data updating method includes: the three-dimensional coordinate errors of multiple data collections can be eliminated by setting permanent image control points around the building and incorporating them into aerial triangulation solutions. After obtaining new damage data, it is determined whether the deviation of the damage center position from the previous inspection is within 3 cm and the damage type is consistent to determine whether it belongs to the previous and subsequent inspection results of the same damage. For new damage, a new data is created, and for old damage, the damage data is updated, and the original data is included in the historical evolution of the damage, thereby completing the update management of the damage data on the three-dimensional model of the building exterior surface.

7. The method for inspecting the apparent damage of exterior building surfaces based on drones according to claim 1 is characterized in that: In the pixel size calculation, For surface damage, the pixel size of the damaged area is obtained by calculating the number of pixels in the surface area. For crack damage, the pixel size of the damaged area is obtained by calculating the maximum value of the cumulative number of pixels in the length direction and the number of pixels in the width direction of the crack using the centerline method.

8. The method for inspecting the surface damage of building exterior surfaces based on drones according to claim 1 is characterized in that: In damage detection and segmentation, the images collected by the drone are cut into 1024×1024 detection units and calculated in parallel. Then, adjacent damages of the same type are merged and the local coordinate system is transformed into the global coordinate system to achieve automatic segmentation of drone images and automatic damage detection.

9. The method for inspecting the surface damage of building exterior surfaces based on drone according to claim 1 is characterized in that: In drone deployment, the damage detection instance segmentation deep learning model collects one or more damage images containing cracks, defects, attachments, rust spots, alkali efflorescence, weathering, decay, and peeling and annotates them as data sets for model training. The trained damage detection instance segmentation deep learning model can automatically identify the damage type according to the input image and perform instance extraction.

10. The method for inspecting the apparent damage of exterior building surfaces based on drones according to claim 1 is characterized in that: In the drone deployment, the material classification neural network model collects and annotates damaged images of one or more types of stone, Taishan bricks, plain bricks, real stone paint, wood veneer, water-brushed stone, terrazzo, and paint as data sets for classification task training, so that the trained material classification neural network model has the function of automatically identifying the type of finishing material according to the input image.

Images