Unmanned aerial vehicle intelligent inspection system based on artificial intelligence

Abstract

The invention provides an unmanned aerial vehicle intelligent inspection system based on artificial intelligence, and belongs to the field of unmanned aerial vehicle control. The system comprises a building information acquisition module for acquiring building facade information of a target building and determining an unmanned aerial vehicle group comprising a first unmanned aerial vehicle and a second unmanned aerial vehicle; the unmanned aerial vehicle configuration module obtains a corresponding external facade detector, configures a first unmanned aerial vehicle and a second unmanned aerial vehicle, and obtains a first inspection unmanned aerial vehicle and a second inspection unmanned aerial vehicle; the inspection scheme generation module is used for generating a unit inspection scheme according to the building facade information, the first inspection unmanned aerial vehicle and the second inspection unmanned aerial vehicle; and the cooperative detection control module is used for controlling the first inspection unmanned aerial vehicle and the second inspection unmanned aerial vehicle to carry out facade detection on the target building and transmitting a detection result to the inspection terminal. The technical problem that in the prior art, the efficiency and accuracy of building facade inspection by an unmanned aerial vehicle are low is solved, and the technical effect of improving the efficiency and accuracy of building facade inspection by the unmanned aerial vehicle is achieved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control, and more particularly to an intelligent inspection system for UAVs based on artificial intelligence. Background Technology

[0002] With the acceleration of urbanization, the number of high-rise buildings has increased dramatically, making the safety inspection of building facades a crucial aspect of building maintenance and safety management. Traditional manual inspection methods suffer from high safety risks, long inspection cycles, and high costs, failing to meet the demands of large-scale building facade inspections. In recent years, drone technology has been widely applied in the field of building inspection. Existing drone inspection technology primarily uses a single drone to acquire images, which are then transmitted to an inspection terminal for analysis and processing. However, this method has limited efficiency. When dealing with large building facades, it requires a significant amount of time for area-by-area inspection, and the data transmission to the inspection terminal for processing results in low efficiency. Furthermore, the lack of an effective verification mechanism, relying solely on single inspection results, easily leads to false positives and false negatives, resulting in low inspection accuracy. Summary of the Invention

[0003] This invention addresses the technical problems of low efficiency and low accuracy in drone inspection of building facades in existing technologies by providing an artificial intelligence-based intelligent drone inspection system.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides an AI-based intelligent drone inspection system, comprising: a building information acquisition module for acquiring building facade information of a target building and determining a drone group to inspect the target building, the drone group including a first drone and a second drone; a drone configuration module for acquiring corresponding facade detectors based on the building facade information and configuring the first drone and the second drone according to the building facade information and the facade detectors, thereby obtaining a first inspection drone and a second inspection drone; an inspection plan generation module for generating a drone group inspection plan based on the building facade information, the first inspection drone, and the second inspection drone; and a collaborative detection control module for controlling the first inspection drone and the second inspection drone to perform facade inspection of the target building according to the drone group inspection plan and transmitting the facade inspection results to an inspection terminal.

[0005] Optionally, the building facade information includes facade material information, and both the first UAV and the second UAV include an image acquisition unit and an edge processing unit. The UAV configuration module is further configured to: extract the corresponding facade detector from the detector repository according to the facade material information, wherein the facade detector has an inspection image information identifier and a verification image information identifier, and the facade detector includes an inspection image processing branch and a verification image processing branch; set an inspection image acquisition window based on the inspection image information identifier, and set a verification image acquisition window based on the verification image information identifier; configure the image acquisition unit of the first UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the first UAV to obtain the first inspection UAV; configure the image acquisition unit of the second UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the second UAV to obtain the second inspection UAV.

[0006] Optionally, the detector repository construction step includes: acquiring multiple facade material types, determining a first facade material type from the facade material types, and setting inspection image constraints and verification image constraints corresponding to the first facade material type; collecting a first facade inspection image set based on the first facade material type and the inspection image constraints, and labeling the first facade inspection image set to obtain a first inspection image label set; training and generating a first inspection image processing branch based on the first facade inspection image set and the first inspection image label set; and processing the first facade inspection image set based on the verification image constraints. The system processes and obtains a first facade verification image set, and annotates the first facade verification image set to obtain a first verification image annotation set. Based on the first facade verification image set and the first verification image annotation set, a first verification image processing branch is trained and generated. The first inspection image processing branch and the first verification image processing branch are integrated into a first facade detector, and stored in the detector repository in association with the first facade material type. The system trains facade detectors for the other facade material types in the same way as it trains the first facade detector corresponding to the first facade material type, thus constructing the detector repository.

[0007] Optionally, the building facade information also includes facade structure information; the inspection plan generation module is further configured to: extract facade structure information from the building facade information, and divide a first inspection area and a second inspection area according to the facade structure information; set the first inspection area as the inspection area of ​​the first inspection drone, and set the second inspection area as the cross-validation area of ​​the first inspection drone to obtain a first drone inspection plan; set the second inspection area as the inspection area of ​​the second inspection drone, and set the first inspection area as the cross-validation area of ​​the second inspection drone to obtain a second drone inspection plan; and generate the unit inspection plan according to the first drone inspection plan and the second drone inspection plan.

[0008] Optionally, the collaborative detection control module is further configured to: control the first inspection drone to traverse and inspect the first inspection area according to the inspection image acquisition window, process the first inspection acquisition image in real time to obtain a first detection result, and send it to the inspection terminal in combination with the inspection window coordinate frame of the first inspection acquisition image; when the first detection result shows an anomaly, traverse and verify the corresponding area of ​​the corresponding first inspection acquisition image according to the verification image acquisition window, process the first verification acquisition image in real time to obtain a first verification result, and send it to the inspection terminal in combination with the verification window coordinate frame of the first verification acquisition image; control the second inspection drone to traverse and inspect the second inspection area according to the inspection image acquisition window, process the second inspection acquisition image in real time to obtain a second detection result, and send it to the inspection terminal in combination with the inspection window coordinate frame of the second inspection acquisition image; when the second detection result shows an anomaly, traverse and verify the corresponding area of ​​the corresponding second inspection acquisition image according to the verification image acquisition window, process the second verification acquisition image in real time to obtain a second verification result, and send it to the inspection terminal in combination with the verification window coordinate frame of the second verification acquisition image.

[0009] Optionally, the collaborative detection control module is further configured to: when the first detection result shows an anomaly, determine the corresponding exterior area of ​​the first inspection acquisition image as a first abnormal sub-region, and send the inspection window coordinate frame of the first abnormal sub-region as the first abnormal window coordinate frame to the second inspection drone for storage; when the second detection result shows an anomaly, determine the corresponding exterior area of ​​the second inspection acquisition image as a second abnormal sub-region, and send the inspection window coordinate frame of the second abnormal sub-region as the second abnormal window coordinate frame to the first inspection drone for storage; when the second inspection drone completes the second inspection area... After the first inspection drone completes the inspection of the first inspection area, it performs traversal verification of the first abnormal sub-region according to the stored first abnormal window coordinate frame and the verification image acquisition window. It processes the first cross-verification image in real time to obtain the first cross-verification result and sends it to the inspection terminal along with the verification window coordinate frame of the first cross-verification image. After the first inspection drone completes the inspection of the first inspection area, it performs traversal verification of the second abnormal sub-region according to the stored second abnormal window coordinate frame and the verification image acquisition window. It processes the second cross-verification image in real time to obtain the second cross-verification result and sends it to the inspection terminal along with the verification window coordinate frame of the second cross-verification image.

[0010] Optionally, the collaborative detection control module is further configured to: control the image acquisition unit of the first inspection drone to perform window-by-window image acquisition of the first inspection area according to the inspection image acquisition window, and obtain a first inspection acquisition image; input the first inspection acquisition image into the edge processing unit of the first inspection drone, wherein the edge processing unit calls the inspection image processing branch in the facade detector for real-time processing to obtain a first detection result; obtain the inspection window coordinate frame corresponding to the first inspection acquisition image, wherein the inspection window coordinate frame contains the coordinate information of four points of the facade area corresponding to the inspection image acquisition window; and send the first detection result and the inspection window coordinate frame to the inspection terminal.

[0011] Optionally, the collaborative detection control module is further configured to: when the first detection result shows an anomaly, determine the corresponding facade area of ​​the first inspection acquisition image as the first abnormal sub-region; control the image acquisition unit of the first inspection drone to perform window-by-window image acquisition of the first abnormal sub-region according to the verification image acquisition window to obtain the first verification acquisition image; input the first verification acquisition image into the edge processing unit of the first inspection drone, the edge processing unit calling the verification image processing branch in the facade detector for real-time processing to obtain the first verification result; obtain the verification window coordinate frame corresponding to the first verification acquisition image, the verification window coordinate frame containing the four-point coordinate information of the facade area corresponding to the verification image acquisition window; and send the first verification result and the verification window coordinate frame to the inspection terminal.

[0012] The beneficial effects of this invention are: The building information acquisition module obtains the building facade information of the target building and determines the drone group to inspect the target building. The drone group includes a first drone and a second drone, providing the hardware foundation and building information support for subsequent dual-drone collaborative operations. The drone configuration module obtains the corresponding facade detectors based on the building facade information and configures the first and second drones according to the building facade information and facade detectors, resulting in the first and second inspection drones. This achieves detection configuration for specific building facade features, ensuring the matching of detectors and inspection objects and improving the accuracy of inspection judgments. The inspection plan generation module generates a drone group inspection plan based on the building facade information, the first inspection drone, and the second inspection drone, realizing inspection planning based on building features and drone performance, and improving the efficiency of inspection operations. The collaborative detection and control module controls the first and second inspection drones to conduct facade inspections of the target building according to the unit's inspection plan, and transmits the facade inspection results to the inspection terminal. This enables intelligent detection operations with dual drones working together and real-time transmission of inspection results, ensuring efficient execution of inspection tasks and timely acquisition of results.

[0013] The above technical solutions enable intelligent inspection, from building information acquisition, drone configuration, inspection plan generation to collaborative detection and control, thereby effectively improving the efficiency and accuracy of drone inspections of building facades. Attached Figure Description

[0014] Figure 1 A schematic diagram of the structure of the AI-based unmanned aerial vehicle (UAV) intelligent inspection system provided by the present invention; Figure 2 This is a flowchart illustrating the process of generating unit inspection plans using the inspection plan generation module provided by the present invention.

[0015] In the attached diagram, the components represented by each number are as follows: Building information acquisition module 11, UAV configuration module 12, inspection plan generation module 13, and collaborative detection and control module 14. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0018] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0019] like Figure 1 As shown, this embodiment of the invention provides an artificial intelligence-based unmanned aerial vehicle (UAV) intelligent inspection system, including: The building information acquisition module 11 is used to acquire the building facade information of the target building and determine the drone group to inspect the target building. The drone group includes a first drone and a second drone.

[0020] Specifically, firstly, the building information acquisition module 11 collects and organizes the building facade information of the target building. The target building refers to a building that requires facade inspection, such as a high-rise building, a super high-rise building, or other buildings requiring regular inspection and maintenance. Building facade information refers to various data describing the characteristics of the target building's exterior surface, providing a basic reference for subsequent inspection work. This building facade information can be obtained through various means, including but not limited to reviewing architectural design drawings, conducting preliminary on-site surveys, and querying historical inspection records. The building facade information includes the target building's facade material information, such as different material types like glass curtain walls, stone curtain walls, metal curtain walls, and concrete facades; and the target building's facade structural information, such as building height, width, and facade area.

[0021] Simultaneously, based on the acquired building facade information, the building information acquisition module 11 determines the configuration of the drone equipment to perform this inspection task and identifies the drone group participating in the target building inspection. Specifically, it adopts a dual-drone combination method, meaning the drone group consists of two devices: a first drone and a second drone. Through dual-drone collaborative operation, the comprehensiveness of inspection coverage and operational efficiency can be improved.

[0022] The building information acquisition module 11 completed the preliminary preparation work for the inspection task, namely, it clarified the basic characteristic information of the inspection object and the equipment combination for performing the inspection, laying the foundation for the entire subsequent inspection.

[0023] The drone configuration module 12 is used to obtain the corresponding facade detector based on the building facade information, and configure the first drone and the second drone according to the building facade information and the facade detector to obtain the first inspection drone and the second inspection drone.

[0024] Specifically, firstly, the UAV configuration module 12 obtains the corresponding facade detectors based on the building facade information provided by the building information acquisition module 11. A facade detector is a specialized detection model trained for specific facade material characteristics, capable of identifying and detecting anomalies of the corresponding facade type. A detector repository is pre-built, storing specialized detectors for different facade material types (such as glass curtain walls, stone curtain walls, metal curtain walls, concrete facades, etc.). Based on the facade material information of the target building, the UAV configuration module 12 extracts matching facade detectors from the detector repository, ensuring that the detection algorithm matches the material characteristics of the object to be detected.

[0025] Subsequently, the UAV configuration module 12 configures the first and second UAVs based on the acquired building facade information and facade detectors. The configuration process includes deploying the facade detectors to the UAV's edge processing unit and configuring the UAV's image acquisition parameters according to the requirements of the facade detectors. Through this configuration process, the original first and second UAVs are transformed into first and second inspection UAVs with specific detection capabilities, enabling them to perform inspection tasks targeting the facade features of the target building.

[0026] The UAV configuration module 12 enables the matching deployment of facade detectors, giving the UAV group targeted facade inspection capabilities and preparing for subsequent inspection plan generation and actual inspection operations.

[0027] The inspection plan generation module 13 is used to generate a drone group inspection plan based on the building facade information, the first inspection drone, and the second inspection drone.

[0028] Specifically, the inspection plan generation module 13 comprehensively considers the building facade information, the first inspection drone, and the second inspection drone to generate a drone group inspection plan. The drone group inspection plan refers to a detailed inspection execution plan formulated for the target building, which clarifies the specific work arrangements for the dual-drone collaboration. Based on the building facade information of the target building, including its structural dimensions and material characteristics, and in conjunction with the first and second inspection drones, the inspection plan generation module 13 generates the drone group inspection plan.

[0029] The drone inspection plan includes the inspection areas and work sequence arrangements for the first and second inspection drones in the drone team, ensuring that the two drones can efficiently and collaboratively complete a comprehensive inspection of the target building's facade. The drone inspection plan fully leverages the advantages of dual-drone collaboration, improving the efficiency and accuracy of inspection operations through task division.

[0030] The inspection plan generation module 13 provides a scientific and reasonable operation guidance plan for the UAV group, ensuring that the inspection task can be carried out in an orderly manner according to the predetermined plan, and laying the foundation for subsequent collaborative detection and control.

[0031] The collaborative detection and control module 14 is used to control the first inspection drone and the second inspection drone to perform facade inspection on the target building according to the unit inspection plan, and transmit the facade inspection results to the inspection terminal.

[0032] Specifically, the collaborative detection and control module 14 performs unified flight control and detection task scheduling for the first and second inspection drones based on the inspection plan generated by the inspection plan generation module 13. It establishes a real-time connection with the two inspection drones via wireless communication and sends detection commands to them according to the detection areas, operation sequences, and other requirements specified in the inspection plan, ensuring that the two drones can collaboratively execute the facade inspection task of the target building according to the inspection plan.

[0033] Meanwhile, the collaborative detection and control module 14 is responsible for collecting and transmitting the facade inspection results. The facade inspection results refer to the inspection data obtained by the first and second inspection drones during the execution of their inspection tasks, including inspection conclusions and corresponding location information. Due to the adoption of an edge computing architecture, image processing and analysis are completed in the edge processing unit of the drones. Therefore, there is no need to transmit the original inspection images; only the processed inspection results need to be transmitted, reducing data transmission volume and latency. The collaborative detection and control module 14 receives inspection data from the two inspection drones in real time and transmits these facade inspection results uniformly to the inspection terminal. The inspection terminal refers to the terminal equipment used to receive, display, and store inspection results, such as a ground control station, mobile terminal, or cloud server.

[0034] Through the collaborative detection and control module 14, unified command and dispatch of the UAV group and centralized management of detection data are realized, ensuring the orderly execution of inspection tasks and timely summarization of detection results. This enables automated and intelligent detection of building facades, improving the efficiency and accuracy of UAV inspections.

[0035] Furthermore, the building facade information includes facade material information; both the first drone and the second drone include an image acquisition unit and an edge processing unit; the drone configuration module 12 is also used for: The corresponding facade detector is extracted from the detector repository based on the facade material information. The facade detector has inspection image information identifier and verification image information identifier. The facade detector includes an inspection image processing branch and a verification image processing branch. An inspection image acquisition window is set based on the inspection image information identifier, and a verification image acquisition window is set based on the verification image information identifier; Configure the image acquisition unit of the first UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the first UAV to obtain the first inspection UAV; Configure the image acquisition unit of the second UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the second UAV to obtain the second inspection UAV.

[0036] In one optional implementation, the building facade information includes facade material information, which details the material composition of the target building facade, such as specific material types like glass curtain walls, stone curtain walls, metal curtain walls, and concrete facades. Both the first and second UAVs are equipped with image acquisition units and edge processing units. The image acquisition unit includes optical devices such as cameras and lens assemblies, responsible for acquiring high-definition image data of the facade; the edge processing unit includes computing devices such as processors and memory, responsible for real-time image processing and detection analysis at the UAV end, realizing edge computing functionality.

[0037] First, the UAV configuration module 12 extracts the corresponding facade detector from a pre-built detector repository based on the facade material information of the target building. The detector repository is a pre-built algorithm model library that stores detection models specifically trained for different facade material types, with each material corresponding to a dedicated facade detector. The facade detector has two types of image size parameter identifiers: inspection image information identifiers and verification image information identifiers. Inspection image information identifiers indicate the image size specifications to be acquired during the routine inspection phase; these are larger sizes used to cover a larger facade area for initial scanning and detection. Verification image information identifiers indicate the image size specifications to be acquired during the anomaly verification phase; these are smaller sizes used for refined analysis of detected anomaly areas. Accordingly, the facade detector's algorithm architecture includes two independent processing branches: an inspection image processing branch and a verification image processing branch. The inspection image processing branch is specifically for processing large-sized inspection images, while the verification image processing branch is specifically for processing small-sized verification images.

[0038] Secondly, the UAV configuration module 12 sets image acquisition parameters based on the image size parameter identifier of the facade detector. Specifically, it sets an inspection image acquisition window based on the inspection image information identifier. The inspection image acquisition window defines the specific parameters for image acquisition during routine inspections, including the image width, height, resolution, and other indicators, to guide the UAV in acquiring images during large-scale preliminary inspections. It also sets a verification image acquisition window based on the verification image information identifier. The verification image acquisition window defines the specific parameters for image acquisition during anomaly area verification. Its image size is smaller than that of the inspection image acquisition window, and it guides the UAV to perform more refined segmentation and detection of detected anomalies, that is, to divide the anomaly area into multiple small areas for precise detection one by one.

[0039] Subsequently, the UAV configuration module 12 performs unified hardware configuration and software deployment for the first and second UAVs. In terms of hardware configuration, based on the parameter requirements of the inspection image acquisition window and the verification image acquisition window, the image acquisition units of the first and second UAVs are parameter-set, including adjusting technical parameters such as camera shooting mode and image resolution, so that they can acquire image data of corresponding specifications according to different acquisition window requirements. In terms of software deployment, the extracted facade detector is deployed to the edge processing units of the first and second UAVs, enabling the UAVs to possess corresponding image processing and detection analysis capabilities. Through the above hardware configuration and software deployment, the original first and second UAVs are transformed into a first inspection UAV and a second inspection UAV with professional detection capabilities.

[0040] Through the above configuration process, the detection algorithm and hardware equipment are accurately matched and optimized, enabling the inspection drone to have a two-level layered detection capability for specific facade materials, namely a detection mode that combines inspection and verification, providing complete technical support for subsequent collaborative inspection operations.

[0041] Furthermore, the steps for building the detector repository include: Obtain multiple facade material types, determine the first facade material type from the facade material types, and set the inspection image constraints and verification image constraints corresponding to the first facade material type; Based on the material type of the first facade and the constraints of the inspection images, a first facade inspection image set is collected, and the first facade inspection image set is annotated to obtain a first inspection image annotation set. Based on the first facade inspection image set and the first inspection image annotation set, a first inspection image processing branch is trained and generated. Based on the verification image constraints, the first facade inspection image set is processed to obtain the first facade verification image set, and the first facade verification image set is annotated to obtain the first verification image annotation set. Based on the first facade verification image set and the first verification image annotation set, a first verification image processing branch is trained and generated. The first inspection image processing branch and the first verification image processing branch are integrated into a first facade detector, and stored in the detector repository in association with the first facade material type; Train facade detectors for the other facade material types in the same way as training the first facade detector corresponding to the first facade material type, and build the detector repository.

[0042] In a preferred embodiment, firstly, various facade material types are acquired, covering common facade materials used in building construction, such as glass curtain walls, stone curtain walls, metal curtain walls, concrete facades, and composite material curtain walls. One facade material type is selected from these types as the first facade material type. Subsequently, corresponding image acquisition constraints are set for this first facade material type, including inspection image constraints and verification image constraints. The inspection image constraints define the size, resolution, and other parameter requirements of the images during the inspection phase in the training data, corresponding to the inspection image information identifier in subsequent applications. The verification image constraints define the size, resolution, and other parameter requirements of the images during the verification phase in the training data, with the image size smaller than the inspection image constraints, corresponding to the verification image information identifier in subsequent applications.

[0043] Then, based on the material type of the first facade and the constraints of the inspection images, a set of first facade inspection images that meets the requirements is collected. This set of first facade inspection images contains a large number of facade image samples of this material type, and the image size and quality meet the requirements of the inspection image constraints. Subsequently, the first facade inspection image set is manually annotated, mainly to identify whether there are defects or anomalies in the images, such as cracks or peeling, resulting in a first inspection image annotation set. Based on the first facade inspection image set and the first inspection image annotation set, a deep learning algorithm is used for supervised learning training to generate a first inspection image processing branch specifically for processing inspection images of this material type of facade. For example, large-size image data from the first facade inspection image set is used as the input layer of the neural network. These images cover a large area of ​​the facade, meeting the needs of large-area scanning and detection in the inspection stage; the anomaly markers (normal / abnormal) in the first inspection image annotation set are used as classification labels for the output layer. A convolutional neural network architecture suitable for facade detection, such as ResNet or YOLO, is used to build a deep learning model. During training, inspection images are input into the network through forward propagation. The network extracts multi-layer feature representations of the images and outputs detection prediction results. The cross-entropy loss function value between the detection prediction results and the true labels is calculated. Then, the gradient of each layer parameter is calculated through backpropagation, and optimization algorithms such as stochastic gradient descent are used to update the network weight parameters. After multiple rounds of iterative training, the visual feature differences of the facade material type in normal and abnormal states are gradually learned. Finally, the first inspection image processing branch is formed, which can accurately identify facade anomalies of this material type in UAV inspection scenarios.

[0044] Next, image processing is performed on the collected first facade inspection image set based on verification image constraints. Specifically, the inspection images in the first facade inspection image set are segmented according to the size requirements of the verification image constraints, resulting in a smaller first facade verification image set, realizing the data conversion from large-size inspection images to small-size verification images. Each small-size image in the first facade verification image set is labeled to indicate whether it contains anomalies, resulting in a first verification image label set. Based on the first facade verification image set and the first verification image label set, a deep learning algorithm is used for training to generate a first verification image processing branch specifically for processing verification images of this material type. For example, the small-size image data in the first facade verification image set is used as the input layer of the neural network. These images focus on local detail areas, meeting the needs of refined detection in the verification stage; the anomaly labels in the first verification image label set are used as classification labels for the output layer. Because the verification images are small in size and focus on extracting detailed features, a more sophisticated convolutional neural network architecture, such as DenseNet or EfficientNet, is used to build a deep learning model for small images. During training, the verification image is input into the network through forward propagation. The network extracts deep features from local details and outputs accurate detection predictions. The loss function between the detection predictions and the true labels is calculated. The network parameters are updated through backpropagation, enabling the model to capture subtle features. After iterative training with a large number of small sample images, the model learns the ability to identify subtle anomalies in local areas. Finally, it converges to form the first verification image processing branch, which can accurately subdivide and detect anomalies in the UAV verification stage.

[0045] Subsequently, the trained first inspection image processing branch and first verification image processing branch are integrated to form a complete dual-branch detection model, namely the first facade detector. This first facade detector has the ability to process both large-size inspection images and small-size verification images. An association is established between the first facade detector and the first facade material type, and stored in the detector repository, establishing a mapping relationship between material types and detectors. Then, following the complete process of training the first facade detector corresponding to the first facade material type, the same training process is performed for each of the remaining facade material types, ultimately generating a set of facade detectors covering all facade material types, completing the comprehensive construction of the detector repository.

[0046] Through the above systematic construction process, the detector repository has formed an algorithm model resource library covering a variety of facade materials and possessing dual-level detection capabilities, providing complete support for subsequent UAV intelligent inspection.

[0047] Furthermore, the building facade information also includes facade structure information; such as... Figure 2 As shown, the inspection plan generation module 13 is also used for: Extract the facade structure information from the building facade information, and divide the first inspection area and the second inspection area according to the facade structure information; The first inspection area is set as the inspection area of ​​the first inspection drone, and the second inspection area is set as the cross-validation area of ​​the first inspection drone to obtain the first drone inspection scheme. The second inspection area is set as the inspection area of ​​the second inspection drone, and the first inspection area is set as the cross-validation area of ​​the second inspection drone to obtain the second drone inspection scheme. The unit inspection plan is generated based on the first UAV inspection plan and the second UAV inspection plan.

[0048] In one feasible implementation, the building facade information also includes facade structure information, which describes the geometric structure of the target building's exterior surface. Specifically, it is a three-dimensional coordinate system established based on the target building facade. This three-dimensional coordinate system establishes a three-dimensional spatial coordinate system by selecting a reference point on the facade as the origin. For example, using the bottom point of the lower left corner of the building facade as the origin, a three-dimensional coordinate system with X, Y, and Z axes is established to record the precise coordinate information of each point on the facade, forming a digital geometric representation of the facade.

[0049] The inspection plan generation module 13 formulates an inspection plan for the drone fleet based on the facade structure information and in conjunction with the first and second inspection drones. First, the module extracts facade structure information from the building facade information to obtain the coordinate system data of the target building's facade, acquiring the overall coordinate range and boundary information of the target building's facade. Based on the facade's coordinate boundaries, the entire facade area is divided into a first inspection area and a second inspection area. For example, it can be equally divided along the geometric center line of the facade, or, for instance, using the vertical center line as the boundary, dividing the facade into left and right areas, ensuring that the two inspection areas are essentially equal in coordinate range and area, thus achieving a balanced task allocation.

[0050] Secondly, the inspection plan generation module 13 assigns inspection tasks to the first inspection drone. The first inspection area is set as the primary inspection area for the first inspection drone, meaning the first drone is responsible for routine inspections within this coordinate range. Simultaneously, the second inspection area is set as the cross-validation area for the first drone; that is, when the second drone detects an anomaly in the second inspection area, the first drone needs to perform cross-validation based on the coordinates of the anomaly point. Through this task assignment, the inspection plan for the first drone is obtained.

[0051] Next, the inspection plan generation module 13 assigns corresponding inspection tasks to the second inspection drone. The second inspection area is set as the primary inspection area for the second inspection drone, meaning the second drone is responsible for performing routine inspections within this coordinate range. Simultaneously, the first inspection area is set as the cross-validation area for the second drone; that is, when the first drone detects an anomaly in its first inspection area, the second drone needs to perform cross-validation based on the coordinates of the anomaly point. Through this task allocation, the inspection plan for the second drone is obtained.

[0052] Subsequently, the inspection plan generation module 13 integrates and coordinates the first UAV inspection plan and the second UAV inspection plan to generate a unified crew inspection plan. The crew inspection plan clarifies the operation arrangements of the two inspection UAVs based on the coordinate system, including the coordinate range of their respective main inspection areas and the coordinate range of cross-verification areas. It establishes a dual-machine collaborative working mechanism based on precise coordinate positioning, realizing a collaborative inspection mode with clear task division and mutual verification.

[0053] By formulating the unit inspection plan, precise area division based on the coordinate system and dual-unit collaborative task allocation were achieved, providing a detailed execution plan for subsequent collaborative detection and control.

[0054] Furthermore, the collaborative detection and control module 14 is also used for: The first inspection drone is controlled to traverse the first inspection area according to the inspection image acquisition window, and the first inspection acquisition image is processed in real time to obtain the first detection result. The result is then sent to the inspection terminal in combination with the inspection window coordinate frame of the first inspection acquisition image. When the first detection result shows an abnormality, the corresponding area of ​​the first inspection acquisition image is traversed and verified according to the verification image acquisition window, the first verification acquisition image is processed in real time to obtain the first verification result, and the result is sent to the inspection terminal in combination with the verification window coordinate frame of the first verification acquisition image. The second inspection drone is controlled to traverse the second inspection area according to the inspection image acquisition window, and the second inspection acquisition image is processed in real time to obtain the second detection result. The result is then sent to the inspection terminal in combination with the inspection window coordinate frame of the second inspection acquisition image. When the second detection result shows an anomaly, the corresponding area of ​​the second inspection acquisition image is traversed and verified according to the verification image acquisition window. The second verification acquisition image is processed in real time to obtain the second verification result, which is then sent to the inspection terminal in combination with the verification window coordinate frame of the second verification acquisition image.

[0055] In a preferred embodiment, the collaborative detection control module 14 controls the first inspection drone to perform window-by-window traversal inspection of the first inspection area according to a preset inspection image acquisition window. During flight, the first inspection drone continuously acquires facade images according to the size parameters of the inspection image acquisition window. Each time a first inspection image is acquired, it is transmitted to its edge processing unit. The edge processing unit calls the inspection image processing branch in the facade detector in real time to process and analyze the first inspection image, obtaining a first detection result (normal / abnormal). The collaborative detection control module 14 sends the first detection result along with the coordinate frame of the inspection window corresponding to the first inspection image to the inspection terminal. The coordinate frame of the inspection window records the precise position information of the first inspection image in the facade coordinate system. When the first detection result indicates an abnormality, the collaborative detection control module 14 immediately initiates a verification detection process. According to the size parameters of the verification image acquisition window, a more refined traversal verification detection is performed on the corresponding area of ​​the corresponding first inspection image (the coordinate frame of the inspection window of the first inspection image corresponding to the abnormality), acquiring a smaller first verification image and transmitting it to its edge processing unit. The edge processing unit calls the verification image processing branch in the facade detector to process the first verification acquisition image in real time, obtaining the first verification result (normal / abnormal). The collaborative detection control module 14 sends the first verification result and the verification window coordinate frame corresponding to the first verification acquisition image to the inspection terminal. The verification window coordinate frame records the precise position information of the first verification acquisition image in the facade coordinate system. Since the size of the verification image acquisition window is smaller than the inspection image acquisition window, the coordinate range of the verification window coordinate frame is included within the range of the corresponding inspection window coordinate frame, realizing precise positioning association from coarse inspection to fine inspection.

[0056] In specific implementation, the collaborative detection control module 14 first divides the first inspection area into multiple regularly arranged inspection windows based on the coordinate range of the first inspection area and the size parameters of the inspection image acquisition window, forming a complete inspection window grid. The first inspection drone flies sequentially to the position of each inspection window according to the preset inspection path, acquires the corresponding first inspection acquisition image at each window position, performs real-time processing through the edge processing unit, and obtains the first detection result. When the first detection result shows normal, the first inspection drone continues to fly to the next inspection window to continue detection; when the first detection result shows abnormal, the collaborative detection control module 14 identifies the inspection window as an abnormal area and initiates the verification detection process. For the inspection window that is detected as abnormal, the collaborative detection control module 14 further subdivides the coverage area of ​​the inspection window into multiple smaller verification windows according to the size parameters of the verification image acquisition window, realizing fine segmentation of the abnormal area and obtaining a verification window grid. The first inspection drone performs window-by-window verification within the abnormal area according to the verification window grid, acquires the first verification acquisition image at the position of each verification window, performs real-time processing, and obtains an accurate first verification result. Simultaneously, the collaborative detection and control module 14 sends the coordinate information of the abnormal inspection window as the abnormal window coordinate frame to the second inspection drone for storage, which will be used for subsequent cross-validation. After completing the verification and detection of the current abnormal inspection window, the first inspection drone continues to fly to the next inspection window and repeats the detection process of image acquisition, processing and analysis, and result judgment. This cycle continues until all inspection windows in the first inspection area have been traversed, completing the comprehensive detection task for the entire first inspection area.

[0057] Simultaneously, the collaborative detection control module 14 controls the second inspection drone to perform window-by-window traversal inspection of the second inspection area according to the preset inspection image acquisition window. During flight, the second inspection drone continuously acquires facade images according to the size parameters of the inspection image acquisition window. Each time a second inspection image is acquired, it is transmitted to its edge processing unit. The edge processing unit calls the inspection image processing branch in the facade detector in real time to process and analyze the second inspection image, obtaining a second detection result (normal / abnormal). The collaborative detection control module 14 sends the second detection result along with the coordinate frame of the corresponding inspection window of the second inspection image to the inspection terminal. When the second detection result indicates an abnormality, the collaborative detection control module 14 immediately initiates the verification detection process. According to the size parameters of the verification image acquisition window, a more refined traversal verification detection is performed on the corresponding area of ​​the corresponding second inspection image (the coordinate frame of the inspection window of the second inspection image corresponding to the abnormality), acquiring a smaller second verification image and transmitting it to its edge processing unit. The edge processing unit calls the verification image processing branch in the facade detector to process the second verification image in real time, obtaining a second verification result (normal / abnormal). The collaborative detection control module 14 sends the second verification result and the coordinate frame of the verification window corresponding to the second verification acquisition image to the inspection terminal. In practice, the second inspection drone performs the same detection process as the first inspection drone in the second inspection area, and the two drones operate in parallel to achieve efficient collaborative inspection.

[0058] Through the above control process, the collaborative detection control module 14 realizes a two-level detection mode with dual-machine synchronization, namely a detection mechanism that combines coarse inspection and fine inspection, ensuring the comprehensiveness and accuracy of the detection, and at the same time realizing the precise positioning of the detection results through coordinate frame information.

[0059] Furthermore, the collaborative detection and control module 14 is also used for: When the first detection result shows an anomaly, the corresponding exterior area of ​​the first inspection image is determined as the first abnormal sub-region, and the inspection window coordinate frame of the first abnormal sub-region is sent as the first abnormal window coordinate frame to the second inspection drone for storage. When the second detection result shows an anomaly, the corresponding exterior area of ​​the second inspection acquisition image is determined as the second abnormal sub-region, and the inspection window coordinate frame of the second abnormal sub-region is sent as the second abnormal window coordinate frame to the first inspection drone for storage. After the second inspection drone completes the inspection of the second inspection area, it traverses and verifies the first abnormal sub-region according to the stored first abnormal window coordinate frame and the verification image acquisition window, processes the first cross-verification image in real time to obtain the first cross-verification result, and sends the verification window coordinate frame of the first cross-verification image to the inspection terminal. After the first inspection drone completes the inspection of the first inspection area, it traverses and verifies the second abnormal sub-region according to the stored second abnormal window coordinate frame and the verification image acquisition window, processes the second cross-verification image in real time to obtain the second cross-verification result, and sends the verification window coordinate frame of the second cross-verification image to the inspection terminal.

[0060] In a preferred embodiment, the collaborative detection control module 14 further implements a dual-machine cross-verification mechanism to improve the reliability of detection results through anomaly information sharing and cross-detection.

[0061] When the first detection result shows an anomaly, the collaborative detection control module 14 determines the corresponding facade area of ​​the first inspection image as the first anomaly sub-region. This first anomaly sub-region refers to the specific area range where an anomaly is detected in the facade coordinate system. The collaborative detection control module 14 uses the inspection window coordinate frame of the first anomaly sub-region as the first anomaly window coordinate frame and sends it to the second inspection drone for local storage via wireless communication, establishing a shared record of the anomaly location. Similarly, when the second detection result shows an anomaly, the collaborative detection control module 14 determines the corresponding facade area of ​​the second inspection image as the second anomaly sub-region and sends the inspection window coordinate frame of the second anomaly sub-region as the second anomaly window coordinate frame to the first inspection drone for storage.

[0062] After the second inspection drone completes its main inspection tasks in the second inspection area, the collaborative detection control module 14 initiates the cross-validation process. Based on the previously stored coordinates of the first anomaly window, the second inspection drone flies to the coordinates of the first anomaly sub-region and performs a detailed traversal verification of the first anomaly sub-region according to the size parameters of the verification image acquisition window. The second inspection drone acquires the first cross-validation image and transmits it to its edge processing unit. The edge processing unit calls the verification image processing branch in the facade detector to process the first cross-validation image in real time, obtaining the first cross-validation result. The collaborative detection control module 14 sends the first cross-validation result and the verification window coordinates of the first cross-validation image to the inspection terminal.

[0063] Similarly, after the first inspection drone completes the main inspection tasks of the first inspection area, the collaborative detection control module 14 initiates the corresponding cross-validation process. Based on the stored coordinate frame of the second anomaly window, the first inspection drone flies to the coordinate position of the second anomaly sub-region and performs traversal verification detection of the second anomaly sub-region according to the verification image acquisition window. The first inspection drone acquires the second cross-validation image, which is processed in real time by the edge processing unit to obtain the second cross-validation result. The collaborative detection control module 14 then sends the second cross-validation result and the verification window coordinate frame of the second cross-validation image to the inspection terminal.

[0064] Through the aforementioned cross-validation mechanism, the collaborative detection control module 14 realizes a dual-machine mutual verification detection mode, that is, the anomalies detected by each UAV will be cross-validated by the other UAV, which improves the accuracy and reliability of anomaly detection and effectively avoids the misjudgment problem that may occur in single-machine detection.

[0065] Furthermore, the collaborative detection and control module 14 is also used for: The image acquisition unit of the first inspection drone is controlled to perform window-by-window image acquisition on the first inspection area according to the inspection image acquisition window to obtain the first inspection acquisition image; The first inspection image is input into the edge processing unit of the first inspection drone. The edge processing unit calls the inspection image processing branch in the facade detector to perform real-time processing and obtain the first detection result. Obtain the coordinate frame of the inspection window corresponding to the first inspection image, wherein the coordinate frame of the inspection window contains the coordinate information of four points of the facade area corresponding to the inspection image acquisition window. The first detection result and the coordinate frame of the inspection window are sent to the inspection terminal.

[0066] In a preferred embodiment, firstly, the collaborative detection control module 14 controls the image acquisition unit of the first inspection drone to systematically acquire images of the first inspection area window by window according to a preset inspection image acquisition window. The first inspection drone acquires images of each window position within the first inspection area sequentially according to the planned window coverage strategy, obtaining one first inspection acquisition image per acquisition operation. Then, the collaborative detection control module 14 inputs the acquired first inspection acquisition images to the edge processing unit of the first inspection drone for real-time analysis. After receiving the first inspection acquisition images, the edge processing unit calls the inspection image processing branch in the pre-deployed facade detector to analyze the first inspection acquisition images, determine whether there are facade anomalies in the first inspection acquisition images, and obtain the first detection result.

[0067] Simultaneously, the collaborative detection and control module 14 acquires the coordinate frame of the inspection window corresponding to the first inspection image. The inspection window coordinate frame is a location identifier established based on the facade coordinate system, containing the coordinate information of four points on the facade area corresponding to the inspection image acquisition window of the first inspection image. Specifically, it contains the precise coordinates of the four vertices of the inspection image acquisition window on the facade, namely the upper left, upper right, lower left, and lower right corners, describing the coverage and location information of the first inspection image in the facade space. Next, the collaborative detection and control module 14 associates and combines the first detection result with the corresponding inspection window coordinate frame, and sends it to the inspection terminal via wireless communication. After receiving the detection result and coordinate frame information, the inspection terminal can accurately understand the detection conclusion and the specific location of any anomalies (if any), achieving visualized display and precise positioning of the detection results. Due to the adoption of an edge computing architecture, image processing and analysis are completed at the first inspection drone end, eliminating the need to transmit raw image data to the inspection terminal. Only the processed detection results and coordinate information are transmitted, reducing communication bandwidth requirements and transmission latency, and improving the system's real-time performance and efficiency.

[0068] Similarly, the collaborative detection control module 14 executes the same control process on the second inspection drone, including controlling the image acquisition unit to acquire images of the second inspection area window by window according to the inspection image acquisition window, inputting the second inspection acquired image into the edge processing unit to call the inspection image processing branch for real-time processing to obtain the second detection result, obtaining the inspection window coordinate frame corresponding to the second inspection acquired image, and sending the second detection result and the inspection window coordinate frame to the inspection terminal.

[0069] Through the above-mentioned refined control process, the collaborative detection control module 14 realizes full-process control of UAV detection operations, ensuring the orderly execution and accuracy of image acquisition, processing and analysis, location identification and result transmission.

[0070] Furthermore, the collaborative detection and control module 14 is also used for: When the first detection result shows an anomaly, the corresponding exterior area of ​​the first inspection image is determined as the first abnormal sub-region. The image acquisition unit of the first inspection drone is controlled to perform window-by-window image acquisition on the first abnormal sub-region according to the verification image acquisition window to obtain the first verification acquisition image; The first verification image is input into the edge processing unit of the first inspection drone. The edge processing unit calls the verification image processing branch in the facade detector to perform real-time processing and obtain the first verification result. Obtain the coordinate frame of the verification window corresponding to the first verification acquisition image. The coordinate frame of the verification window contains the coordinate information of four points of the exterior area corresponding to the verification image acquisition window. The first verification result and the coordinate frame of the verification window are sent to the inspection terminal.

[0071] In a preferred embodiment, when the first detection result indicates an anomaly, the collaborative detection control module 14 immediately identifies the corresponding facade area of ​​the first inspection image as the first anomaly sub-region. The first anomaly sub-region refers to the specific facade area within the coverage of the first inspection image where an anomaly is detected. This area requires more refined verification detection to confirm the authenticity and specific characteristics of the anomaly. Then, the collaborative detection control module 14 controls the image acquisition unit of the first inspection drone to reposition itself to the first anomaly sub-region and performs window-by-window image acquisition of the first anomaly sub-region according to a preset verification image acquisition window. Since the size of the verification image acquisition window is smaller than the inspection image acquisition window, the first inspection drone can perform more refined segmentation and acquisition of the first anomaly sub-region, obtaining multiple smaller but higher-resolution first verification acquisition images, achieving refined coverage of the anomaly area.

[0072] Subsequently, the collaborative detection and control module 14 inputs the acquired first verification image to the edge processing unit of the first inspection drone for verification analysis. The edge processing unit calls the verification image processing branch in the facade detector. This branch is specifically optimized for small-sized verification images, enabling it to identify more subtle features and process the first verification image in real time, resulting in a more accurate first verification result. Next, the collaborative detection and control module 14 acquires the verification window coordinate frame corresponding to the first verification image. The verification window coordinate frame is also based on the facade coordinate system and contains the coordinate information of four points on the facade area corresponding to the verification image acquisition window. That is, the precise coordinate values ​​of the four vertices of the verification acquisition window on the facade, namely the upper left, upper right, lower left, and lower right corners. Its coordinate range is within the corresponding inspection window coordinate frame, realizing precise positioning association from coarse inspection to fine inspection. Afterwards, the collaborative detection and control module 14 associates and combines the first verification result with the corresponding verification window coordinate frame and sends it to the inspection terminal via wireless communication. The inspection terminal can receive the detailed verification detection results and precise anomaly location information, realizing precise positioning and detailed analysis of the anomaly area. Also employing an edge computing architecture, image processing during the verification phase is completed on the drone, maintaining the system's real-time performance and efficiency.

[0073] Through the above verification and testing process, the collaborative detection control module 14 realizes a two-level detection mechanism from coarse inspection to fine inspection. When an anomaly is found in the preliminary inspection, it can automatically start fine verification to ensure the accuracy and reliability of the test results.

[0074] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0075] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0080] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI-based unmanned aerial vehicle (UAV) intelligent inspection system, characterized in that: The system includes: The building information acquisition module is used to acquire the building facade information of the target building and determine the drone group to inspect the target building. The drone group includes a first drone and a second drone. The drone configuration module is used to obtain the corresponding facade detector based on the building facade information, and configure the first drone and the second drone according to the building facade information and the facade detector to obtain the first inspection drone and the second inspection drone. The inspection plan generation module is used to generate a drone group inspection plan based on the building facade information, the first inspection drone, and the second inspection drone. The collaborative detection and control module is used to control the first inspection drone and the second inspection drone to perform facade inspection on the target building according to the unit inspection plan, and transmit the facade inspection results to the inspection terminal.

2. The system according to claim 1, characterized in that, The building facade information includes facade material information; both the first drone and the second drone include an image acquisition unit and an edge processing unit; the drone configuration module is further used for: The corresponding facade detector is extracted from the detector repository based on the facade material information. The facade detector has inspection image information identifier and verification image information identifier. The facade detector includes an inspection image processing branch and a verification image processing branch. An inspection image acquisition window is set based on the inspection image information identifier, and a verification image acquisition window is set based on the verification image information identifier; Configure the image acquisition unit of the first UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the first UAV to obtain the first inspection UAV; Configure the image acquisition unit of the second UAV according to the inspection image acquisition window and the verification image acquisition window, and deploy the facade detector in the edge processing unit of the second UAV to obtain the second inspection UAV.

3. The system according to claim 2, characterized in that, The steps for building the detector repository include: Obtain multiple facade material types, determine the first facade material type from the facade material types, and set the inspection image constraints and verification image constraints corresponding to the first facade material type; Based on the material type of the first facade and the constraints of the inspection images, a first facade inspection image set is collected, and the first facade inspection image set is annotated to obtain a first inspection image annotation set. Based on the first facade inspection image set and the first inspection image annotation set, a first inspection image processing branch is trained and generated. Based on the verification image constraints, the first facade inspection image set is processed to obtain the first facade verification image set, and the first facade verification image set is annotated to obtain the first verification image annotation set. Based on the first facade verification image set and the first verification image annotation set, a first verification image processing branch is trained and generated. The first inspection image processing branch and the first verification image processing branch are integrated into a first facade detector, and stored in the detector repository in association with the first facade material type; Train facade detectors for the other facade material types in the same way as training the first facade detector corresponding to the first facade material type, and build the detector repository.

4. The system according to claim 2, characterized in that, The building facade information also includes facade structure information; the inspection plan generation module is also used for: Extract the facade structure information from the building facade information, and divide the first inspection area and the second inspection area according to the facade structure information; The first inspection area is set as the inspection area of ​​the first inspection drone, and the second inspection area is set as the cross-validation area of ​​the first inspection drone to obtain the first drone inspection scheme. The second inspection area is set as the inspection area of ​​the second inspection drone, and the first inspection area is set as the cross-validation area of ​​the second inspection drone to obtain the second drone inspection scheme. The unit inspection plan is generated based on the first UAV inspection plan and the second UAV inspection plan.

5. The system according to claim 4, characterized in that, The collaborative detection and control module is also used for: The first inspection drone is controlled to traverse the first inspection area according to the inspection image acquisition window, and the first inspection acquisition image is processed in real time to obtain the first detection result. The result is then sent to the inspection terminal in combination with the inspection window coordinate frame of the first inspection acquisition image. When the first detection result shows an abnormality, the corresponding area of ​​the first inspection acquisition image is traversed and verified according to the verification image acquisition window, the first verification acquisition image is processed in real time to obtain the first verification result, and the result is sent to the inspection terminal in combination with the verification window coordinate frame of the first verification acquisition image. The second inspection drone is controlled to traverse the second inspection area according to the inspection image acquisition window, and the second inspection acquisition image is processed in real time to obtain the second detection result. The result is then sent to the inspection terminal in combination with the inspection window coordinate frame of the second inspection acquisition image. When the second detection result shows an anomaly, the corresponding area of ​​the second inspection acquisition image is traversed and verified according to the verification image acquisition window. The second verification acquisition image is processed in real time to obtain the second verification result, which is then sent to the inspection terminal in combination with the verification window coordinate frame of the second verification acquisition image.

6. The system according to claim 5, characterized in that, The collaborative detection and control module is also used for: When the first detection result shows an anomaly, the corresponding exterior area of ​​the first inspection image is determined as the first abnormal sub-region, and the inspection window coordinate frame of the first abnormal sub-region is sent as the first abnormal window coordinate frame to the second inspection drone for storage. When the second detection result shows an anomaly, the corresponding exterior area of ​​the second inspection acquisition image is determined as the second abnormal sub-region, and the inspection window coordinate frame of the second abnormal sub-region is sent as the second abnormal window coordinate frame to the first inspection drone for storage. After the second inspection drone completes the inspection of the second inspection area, it performs traversal verification of the first abnormal sub-region according to the stored first abnormal window coordinate frame and the verification image acquisition window, processes the first cross-verification image in real time to obtain the first cross-verification result, and sends the verification window coordinate frame of the first cross-verification image to the inspection terminal. After the first inspection drone completes the inspection of the first inspection area, it traverses and verifies the second abnormal sub-region according to the stored second abnormal window coordinate frame and the verification image acquisition window, processes the second cross-verification image in real time to obtain the second cross-verification result, and sends the verification window coordinate frame of the second cross-verification image to the inspection terminal.

7. The system according to claim 5, characterized in that, The collaborative detection and control module is also used for: The image acquisition unit of the first inspection drone is controlled to perform window-by-window image acquisition on the first inspection area according to the inspection image acquisition window to obtain the first inspection acquisition image; The first inspection image is input into the edge processing unit of the first inspection drone. The edge processing unit calls the inspection image processing branch in the facade detector to perform real-time processing and obtain the first detection result. Obtain the coordinate frame of the inspection window corresponding to the first inspection image, wherein the coordinate frame of the inspection window contains the coordinate information of four points of the facade area corresponding to the inspection image acquisition window. The first detection result and the coordinate frame of the inspection window are sent to the inspection terminal.

8. The system according to claim 5, characterized in that, The collaborative detection and control module is also used for: When the first detection result shows an anomaly, the corresponding exterior area of ​​the first inspection image is determined as the first abnormal sub-region. The image acquisition unit of the first inspection drone is controlled to perform window-by-window image acquisition on the first abnormal sub-region according to the verification image acquisition window to obtain the first verification acquisition image; The first verification image is input into the edge processing unit of the first inspection drone. The edge processing unit calls the verification image processing branch in the facade detector to perform real-time processing and obtain the first verification result. Obtain the coordinate frame of the verification window corresponding to the first verification acquisition image. The coordinate frame of the verification window contains the coordinate information of four points of the exterior area corresponding to the verification image acquisition window. The first verification result and the coordinate frame of the verification window are sent to the inspection terminal.

Images