A large-space drone automatic inspection method and system based on laser radar

Abstract

The present application discloses a method and system for automatic inspection of large-space UAVs based on laser radar, which belongs to the field of computer technology. The method obtains the original image collected by the image acquisition component of the UAV, the original point cloud collected by the laser radar, the first environmental information of the environment in which the UAV is located, the second environmental information of the inspection position of the UAV, and the dynamic acquisition parameters. Based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters, the environmental image correction information and the environmental point cloud correction information are determined. Based on the inspection indication information, the inspection position, the environmental image correction information, and the environmental point cloud correction information, the image enhancement parameters and the point cloud enhancement parameters are determined. Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain the target image and the target point cloud. Based on the target image and the target point cloud, it is determined whether there is an abnormality at the inspection position.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a laser radar-based large-space unmanned aerial vehicle automatic inspection method and system. Background Art

[0002] With the widespread application of drones in agricultural monitoring, disaster assessment, geographic surveying and mapping, etc., a large number of aerial images can be obtained by using drones for shooting, and aerial images can be used to realize inspections of specific spaces.

[0003] However, aerial images are greatly affected by factors such as weather. In the face of special weather conditions, the effect of using aerial images to inspect specific spaces is poor, that is, abnormalities in specific spaces cannot be discovered in a timely manner. Summary of the Invention

[0004] The present application provides a method and system for automated inspection of large spaces by drones based on lidar, which can improve the effectiveness of inspecting specific spaces using drones and promptly detect anomalies in the specific spaces. The technical solution is as follows:

[0005] On the one hand, a large-space UAV automatic inspection method based on laser radar is provided, the method comprising:

[0006] Obtaining an original image captured by an image acquisition component of a drone, an original point cloud captured by a lidar, first environmental information of an environment in which the drone is located, second environmental information of an inspection location of the drone, and dynamic acquisition parameters, the dynamic acquisition parameters including inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters;

[0007] Determining environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters;

[0008] Determining image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information;

[0009] Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain a target image and a target point cloud;

[0010] Based on the target image and the target point cloud, it is determined whether there is any abnormality in the inspection position.

[0011] On the one hand, a large-space UAV automatic inspection system based on laser radar is provided, and the device includes:

[0012] An acquisition module is configured to acquire an original image captured by the image acquisition component of the drone, an original point cloud captured by the lidar, first environmental information of the drone's environment, second environmental information of the drone's inspection location, and dynamic acquisition parameters, wherein the dynamic acquisition parameters include inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters;

[0013] a correction information determination module, configured to determine environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters;

[0014] an enhancement parameter determination module, configured to determine image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information;

[0015] a processing module, configured to process the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters, respectively, to obtain a target image and a target point cloud;

[0016] The abnormality recognition module is used to determine whether there is an abnormality in the inspection position based on the target image and the target point cloud.

[0017] On the one hand, a computer device is provided, which includes one or more processors and one or more memories, wherein at least one computer program is stored in the one or more memories, and the computer program is loaded and executed by the one or more processors to implement the large-space drone automatic inspection method based on lidar.

[0018] On the one hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The computer program is loaded and executed by a processor to implement the large-space UAV automatic inspection method based on laser radar.

[0019] On the one hand, a computer program product or computer program is provided, which includes a program code, which is stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium, and the processor executes the program code, so that the computer device executes the above-mentioned large-space drone automatic inspection method based on lidar.

[0020] Through the technical solution provided in the embodiment of the present application, the original image collected by the image acquisition component of the UAV, the original point cloud collected by the lidar, the first environmental information of the environment in which the UAV is located, the second environmental information of the inspection position of the UAV and the dynamic acquisition parameters are obtained. Based on the first environmental information, the second environmental information, the dynamic image acquisition parameters and the dynamic point cloud acquisition parameters, the environmental image correction information and the environmental point cloud correction information are determined. Based on the inspection indication information, the inspection position, the environmental image correction information and the environmental point cloud correction information, the image enhancement parameters and the point cloud enhancement parameters are determined. Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain the target image and the target point cloud. Based on the target image and the target point cloud, it is determined whether there is an abnormality at the inspection position. Multi-dimensional information is combined during the inspection, the inspection effect is higher, and anomalies in a specific space can be discovered in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 Schematic diagram of an implementation environment of a large-space UAV automatic inspection method based on laser radar provided in an embodiment of the present application;

[0023] Figure 2 This is a flow chart of a large-space UAV automatic inspection method based on laser radar provided in an embodiment of the present application;

[0024] Figure 3 This is a flow chart of another large-space UAV automatic inspection method based on lidar provided in an embodiment of the present application;

[0025] Figure 4 This is a schematic diagram of the structure of a large-space UAV automatic inspection system based on laser radar provided in an embodiment of the present application;

[0026] Figure 5 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0028] In this application, the terms "first", "second", etc. are used to distinguish identical or similar items with substantially the same effects and functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor is there any limitation on the quantity and execution order.

[0029] UAV: The abbreviation of Unmanned Aerial Vehicle (UAV), is an unmanned aerial vehicle that is controlled by radio remote control equipment and its own program control device.

[0030] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to achieve better results.

[0031] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge sub-models to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning through demonstration.

[0032] Normalization: Mapping sequences of numbers with different value ranges to the interval (0, 1) facilitates data processing. In some cases, the normalized values ​​can be directly implemented as probabilities.

[0033] Embedded Coding: Mathematically, embedded coding represents a correspondence relationship, mapping data in X space to Y space using a function F, where F is an injective function. The mapping result preserves structure. An injective function indicates that the data after mapping uniquely corresponds to the data before mapping. Structural preservation means that the size relationship of the data before mapping is the same as the size relationship of the data after mapping. For example, if there are data X1 and X2 before mapping, the data Y1 corresponding to X1 and Y2 corresponding to X2 will be obtained after mapping. If the data X1 before mapping is greater than X2, then the data Y1 after mapping is greater than Y2. For words, this means mapping the words to another space to facilitate subsequent machine learning and processing.

[0034] Attention weight: This indicates the importance of a piece of data during training or prediction. Importance indicates the impact of the input data on the output data. Highly important data has a higher attention weight, while lowly important data has a lower attention weight. Data importance varies in different scenarios, and the process of training a model's attention weights is also the process of determining data importance.

[0035] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In addition, the shooting locations and inspection objects of the drones in the embodiments of this application are all shooting locations and inspection objects that are permitted to be photographed by drones by relevant laws and regulations.

[0036] Figure 1 This is a schematic diagram of the implementation environment of an image data analysis method based on an AI model provided in an embodiment of the present application, see Figure 1 The implementation environment may include a drone 110, a ground information collection device 120 and a server 140.

[0037] Drone 110 is connected to server 140 via a wireless network and can transmit collected information to server 140. Drone 110 includes an image acquisition component and an environmental information acquisition component. The image acquisition component is used to acquire images, and the environmental information acquisition component is used to acquire environmental information. In this embodiment of the present application, the image acquisition component uses dynamic image acquisition parameters when acquiring images, thereby accommodating a wider range of shooting scenarios.

[0038] The ground information collection device 120 is installed on the ground within the shooting area of ​​the drone 110. The shooting area is also the area captured by the drone 110. There are multiple ground information collection devices 120 in the shooting area, and different ground information collection devices 120 can collect relevant information from different shooting locations. The ground information collection devices 120 are connected to the server 140 via a wireless network and can send the collected information to the server 140.

[0039] Server 140 is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, a Content Delivery Network (CDN), and big data and artificial intelligence platforms. Server 140 can provide background services for applications running on drone 110 and ground information collection equipment 120, thereby executing the steps provided in the embodiments of this application to achieve the purpose of the technical solutions provided in the embodiments of this application.

[0040] After introducing the implementation environment of the embodiment of the present application, the application scenario of the technical solution provided by the embodiment of the present application is described below. The technical solution provided by the embodiment of the present application is applied in the scenario of analyzing images collected by drones, and the implementation of the technical solution provided by the embodiment of the present application requires the cooperation of the above-mentioned ground information acquisition equipment 120. The ground information acquisition equipment 120 can be a fixed acquisition equipment or a collection equipment that can be disassembled and moved. For example, for a specific space, multiple ground information acquisition devices can be arranged in the specific space in advance, and then drones can be used to perform image acquisition, point cloud data and environmental information acquisition, and the information collected by the drones and the information collected by the ground information acquisition equipment 120 can be used to realize inspections of specific spaces, thereby improving the inspection effect.

[0041] After introducing the implementation environment and application scenarios of this application, the technical solutions provided by the embodiments of this application are introduced below. Figure 2 Taking the execution subject as a server as an example, the method includes the following steps.

[0042] 201. The server obtains the original image captured by the image acquisition component of the drone, the original point cloud captured by the lidar, the first environmental information of the drone's environment, the second environmental information of the drone's inspection position, and dynamic acquisition parameters, which include inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters.

[0043] Among them, the drone is a drone used to capture images of a specific space. The specific space is a space that is allowed to be captured by drones by relevant laws and regulations. For example, the specific space is farmland space, forest space, park space, and nature reserve space. The use of drones can achieve efficient inspections of specific spaces, and the specific space is also a large space. The original image is the image directly captured by the drone, that is, the aerial image. In the actual working process, the number of original images captured by the drone is large. For the sake of convenience, the embodiment of the present application is described with the number of original images as one, and the processing methods of other original images all belong to the same inventive concept. The original point cloud is the point cloud data directly collected by the drone through the lidar, and the original point cloud is collected at the same time as the original image. The first environmental information is the environmental information of the environment in which the drone is located, that is, the environmental information of the position where the drone is located when the original image is captured. Generally speaking, the first environmental information is the environmental information in the air. The inspection position is the shooting position corresponding to the original image. The inspection position is a ground position. The second environmental information is the environmental information collected by the ground information collection device. The ground information collection device is the ground information collection device closest to the shooting position in the specific space. The second environmental information is the ground environmental information when the drone shoots the shooting position. The inspection indication information, dynamic image acquisition parameters and dynamic point cloud acquisition parameters in the dynamic acquisition parameters are all dynamically changeable. The inspection indication information includes the inspection target and the inspection strategy. The inspection target is the purpose of using the drone to inspect a specific space. For example, when the inspection area is a farmland area, the inspection target can be to determine whether there are insect pests in the farmland area; when the inspection area is a forest area, the inspection target can be to determine whether there are fire hazards in the forest. The inspection target is configured by technicians according to actual conditions. In the embodiment of the present application, the inspection indication information is in the form of natural language. The dynamic image acquisition parameters are determined based on the object attributes of the inspection object, the initial image acquisition parameters of the image acquisition component of the drone, and the flight parameters of the drone. The object properties of the inspection object are used to represent the object characteristics of the inspection object. The inspection object is the object present at the inspection location, for example, the inspection object is the crops or animals at the inspection location. The object properties include the object type, object size, and object surface material. The initial image acquisition parameters are the image acquisition parameters directly determined by the image acquisition component, and are not the image acquisition parameters ultimately used to capture the original image. The dynamic image acquisition parameters are the image acquisition parameters ultimately used to capture the original image. The flight parameters of the drone are used to represent the flight status of the drone. For example, flight parameters include flight altitude and flight speed. The dynamic point cloud acquisition parameters are determined based on the object properties of the inspection object, the initial point cloud acquisition parameters of the drone's lidar, and the flight parameters of the drone.The initial point cloud acquisition parameters of the laser radar of the drone are the point cloud acquisition parameters adopted by the laser radar by default, and are not the point cloud acquisition parameters ultimately used to acquire the original point cloud. The dynamic point cloud acquisition parameters are the point cloud acquisition parameters ultimately used to acquire the original point cloud. The dynamic image acquisition parameters are determined by combining the object properties of the inspection object in the inspection position, the initial image acquisition parameters of the image acquisition component, and the flight parameters of the drone, that is, they are comprehensive parameters determined by combining the actual state of the inspection object, the image acquisition component, and the drone. The dynamic image acquisition parameters are very important parameters in the technical solution provided by the embodiment of the present application and are an important part of the inventive concept of the technical solution provided by the embodiment of the present application. Accordingly, the dynamic point cloud acquisition parameters are determined by combining the object properties of the inspection object in the inspection position, the initial point cloud acquisition parameters of the laser radar, and the flight parameters of the drone, that is, they are comprehensive parameters determined by combining the actual state of the inspection object, the laser radar, and the drone. The dynamic point cloud acquisition parameters are very important parameters in the technical solution provided by the embodiment of the present application and are an important part of the inventive concept of the technical solution provided by the embodiment of the present application.

[0044] 202. The server determines environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters.

[0045] Among them, the environmental image correction information is a parameter used to correct the original image, and the environmental point cloud correction information is a parameter used to correct the original point cloud. After correction, more accurate image and point cloud data can be obtained.

[0046] 203. The server determines image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection location, the environmental image correction information, and the environmental point cloud correction information.

[0047] Among them, the image enhancement parameters are parameters used to enhance the original image, and the point cloud enhancement parameters are parameters used to enhance the original point cloud. Both the image enhancement parameters and the point cloud enhancement parameters are enhancement parameters obtained by integrating multi-dimensional information.

[0048] 204. The server processes the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters to obtain a target image and a target point cloud.

[0049] Among them, the target image and target point cloud are obtained after enhancement. Compared with the original image and original point cloud, they have more information and higher accuracy.

[0050] 205. The server determines whether there is any abnormality at the inspection location based on the target image and the target point cloud.

[0051] Through the technical solution provided in the embodiment of the present application, the original image collected by the image acquisition component of the UAV, the original point cloud collected by the lidar, the first environmental information of the environment in which the UAV is located, the second environmental information of the inspection position of the UAV and the dynamic acquisition parameters are obtained. Based on the first environmental information, the second environmental information, the dynamic image acquisition parameters and the dynamic point cloud acquisition parameters, the environmental image correction information and the environmental point cloud correction information are determined. Based on the inspection indication information, the inspection position, the environmental image correction information and the environmental point cloud correction information, the image enhancement parameters and the point cloud enhancement parameters are determined. Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain the target image and the target point cloud. Based on the target image and the target point cloud, it is determined whether there is an abnormality at the inspection position. Multi-dimensional information is combined during the inspection, the inspection effect is higher, and anomalies in a specific space can be discovered in a timely manner.

[0052] The above steps 201-205 are a brief introduction to the large space drone automatic inspection method based on laser radar provided in the embodiment of the present application. The following will combine some examples to more clearly explain the large space drone automatic inspection method based on laser radar provided in the embodiment of the present application. Figure 3 Taking the execution subject as a server as an example, the method includes the following steps.

[0053] 301. The server obtains the original image captured by the image acquisition component of the drone, the original point cloud captured by the lidar, the first environmental information of the drone's environment, the second environmental information of the drone's inspection position, and dynamic acquisition parameters, which include inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters.

[0054] Among them, the drone is a drone used to capture images of a specific space. The specific space is a space that is allowed to be captured by drones by relevant laws and regulations. For example, the specific space is farmland space, forest space, park space, and nature reserve space. The use of drones can achieve efficient inspections of specific spaces, and the specific space is also a large space. The original image is the image directly captured by the drone, that is, the aerial image. In the actual working process, the number of original images captured by the drone is large. For the sake of convenience, the embodiment of the present application is described with the number of original images as one, and the processing methods of other original images all belong to the same inventive concept. The original point cloud is the point cloud data directly collected by the drone through the lidar, and the original point cloud is collected at the same time as the original image. The first environmental information is the environmental information of the environment in which the drone is located, that is, the environmental information of the position where the drone is located when the original image is captured. Generally speaking, the first environmental information is the environmental information in the air. The inspection position is the shooting position corresponding to the original image. The inspection position is a ground position. The second environmental information is the environmental information collected by the ground information collection device. The ground information collection device is the ground information collection device closest to the shooting position in the specific space. The second environmental information is the ground environmental information when the drone shoots the shooting position. The inspection indication information, dynamic image acquisition parameters and dynamic point cloud acquisition parameters in the dynamic acquisition parameters are all dynamically changeable. The inspection indication information includes the inspection target and the inspection strategy. The inspection target is the purpose of using the drone to inspect a specific space. For example, when the inspection area is a farmland area, the inspection target can be to determine whether there are insect pests in the farmland area; when the inspection area is a forest area, the inspection target can be to determine whether there are fire hazards in the forest. The inspection target is configured by technicians according to actual conditions. In the embodiment of the present application, the inspection indication information is in the form of natural language. The dynamic image acquisition parameters are determined based on the object attributes of the inspection object, the initial image acquisition parameters of the image acquisition component of the drone, and the flight parameters of the drone. The object properties of the inspection object are used to represent the object characteristics of the inspection object. The inspection object is the object present at the inspection location, for example, the inspection object is the crops or animals at the inspection location. The object properties include the object type, object size, and object surface material. The initial image acquisition parameters are the image acquisition parameters directly determined by the image acquisition component, and are not the image acquisition parameters ultimately used to capture the original image. The dynamic image acquisition parameters are the image acquisition parameters ultimately used to capture the original image. The flight parameters of the drone are used to represent the flight status of the drone. For example, flight parameters include flight altitude and flight speed. The dynamic point cloud acquisition parameters are determined based on the object properties of the inspection object, the initial point cloud acquisition parameters of the drone's lidar, and the flight parameters of the drone.The initial point cloud acquisition parameters of the laser radar of the drone are the point cloud acquisition parameters adopted by the laser radar by default, and are not the point cloud acquisition parameters ultimately used to acquire the original point cloud. The dynamic point cloud acquisition parameters are the point cloud acquisition parameters ultimately used to acquire the original point cloud. The dynamic image acquisition parameters are determined by combining the object properties of the inspection object in the inspection position, the initial image acquisition parameters of the image acquisition component, and the flight parameters of the drone, that is, they are comprehensive parameters determined by combining the actual state of the inspection object, the image acquisition component, and the drone. The dynamic image acquisition parameters are very important parameters in the technical solution provided by the embodiment of the present application and are an important part of the inventive concept of the technical solution provided by the embodiment of the present application. Accordingly, the dynamic point cloud acquisition parameters are determined by combining the object properties of the inspection object in the inspection position, the initial point cloud acquisition parameters of the laser radar, and the flight parameters of the drone, that is, they are comprehensive parameters determined by combining the actual state of the inspection object, the laser radar, and the drone. The dynamic point cloud acquisition parameters are very important parameters in the technical solution provided by the embodiment of the present application and are an important part of the inventive concept of the technical solution provided by the embodiment of the present application.

[0055] In some embodiments, the original image is captured by an image acquisition component of a drone. The image acquisition component refers to hardware with image acquisition capabilities, and the image acquisition parameters of the image acquisition component are adjustable. After the drone captures the original image via the image acquisition component, it transmits the original image to the server via a network connection with the server, and the server obtains the original image.

[0056] In some embodiments, the raw point cloud is collected by a drone's lidar. Lidar refers to hardware capable of collecting point clouds, and the lidar's point cloud collection parameters are adjustable. After the drone collects the raw point cloud via the lidar, it transmits the raw point cloud to the server via a network connection, and the server retrieves the raw point cloud.

[0057] In some embodiments, the first environmental information is collected by an environmental sensor on a drone. For example, the first environmental information includes a first ambient light parameter, a first ambient temperature, a first ambient humidity, and a weather type. The first ambient light parameter includes parameters related to ambient light, for example, the first ambient light parameter includes a first ambient light intensity, a first light color temperature, a first light direction, and a first light color. The concepts in the first ambient light parameter are all commonly used concepts in the relevant art. The environmental sensor is actually an ambient light sensor. The ambient light sensor can also use a sensor in the relevant art. The embodiment of the present application does not limit this. In the technical solution provided in the embodiment of the present application, the focus is on processing the collected data. As for the method of obtaining the corresponding data, those skilled in the art can adopt an appropriate method to implement it. After collecting the first environmental information, the first environmental information is sent to the server via a network connection with the server, and the server obtains the first environmental information. The first environmental information and the original image are both collected by the drone. The collection time of the first environmental information is the same as the collection time of the original image. That is, the first environmental information is the environmental information collected at the same time as the original image.

[0058] In some embodiments, the second environmental information is collected by a ground information collection device. For example, the second environmental information includes a second environmental light parameter, a second environmental temperature, and a second environmental humidity. The concepts in the second environmental light parameter are all commonly used concepts in the relevant art. The ground information collection device includes an environmental sensor, through which the second environmental information can be collected. The environmental sensor is actually an ambient light sensor. The ambient light sensor can also use a sensor in the relevant art. The embodiments of the present application are not limited to this. The second environmental information and the first environmental information are environmental information collected by different hardware. In order to maintain the correspondence between the second environmental information and the first environmental information, the timestamp when the environmental information was collected can be used to find the second environmental information corresponding to the first environmental information. For example, the ground information collection device periodically sends the collected environmental information to the server, and the environmental information sent carries the timestamp of the collection time. When the drone sends the first environmental information to the server, it carries the timestamp corresponding to the first environmental information. The server uses the timestamp corresponding to the first environmental information to find the corresponding second environmental information from the environmental information sent by the ground information collection device. The timestamp corresponding to the second environmental information is the same as that of the first environmental information, or the second environmental information is the second environmental information whose timestamp is closest to the timestamp of the first environmental information.

[0059] In an embodiment of the present application, the dynamic image acquisition parameters are determined based on the object attributes of the inspection object, the initial image acquisition parameters of the image acquisition component of the drone, and the flight parameters of the drone. The method for determining the dynamic image acquisition parameters is described below.

[0060] In some embodiments, a controller of the drone obtains object attributes of the inspection object from a server. The controller determines image acquisition parameter correction parameters based on the object attributes and the flight parameters. The controller corrects the initial image acquisition parameters based on the image acquisition parameter correction parameters to obtain the dynamic image acquisition parameters.

[0061] Among them, the drone will transmit the initial image collected in real time to the server. Here, the initial image collected in real time is collected by the drone's image acquisition component based on the initial image acquisition parameters. The server obtains the initial image, performs target detection on the initial image, and obtains the object attributes of the inspection object in the initial image. The server returns the object attributes to the drone's controller, and the drone's controller obtains the object attributes of the inspection object. The difference between the initial image and the original image is that the initial image is collected based on the initial image acquisition parameters, and the original image is collected based on the dynamic image acquisition parameters. The way the server identifies the object attributes of the inspection object in the initial image can adopt the target detection and target recognition methods in the relevant technology, or the server can identify the object type of the inspection object from the initial image, and then use the object type to query the object attributes of the inspection object in the object attribute database. The embodiment of the present application does not limit this. The image acquisition parameter correction coefficient is used to correct the initial image acquisition parameters to obtain the dynamic correction coefficient.

[0062] For example, the drone's controller obtains the object attributes of the inspection object from a server. The controller extracts features from the object attributes and the flight parameters to obtain object attribute features of the object attributes and flight parameter features of the flight parameters. The controller fuses the object attribute features and the flight parameter features to obtain a first correction parameter determination feature. The controller fully connects and normalizes the first correction parameter determination feature to obtain the image acquisition parameter correction parameter. The controller uses the image acquisition parameter correction parameter to correct the initial image acquisition parameters to obtain the dynamic image acquisition parameters.

[0063] Among them, the initial image acquisition parameters include the initial aperture size, the initial shutter speed, the initial sensitivity and the initial white balance, and the image acquisition parameter correction parameters include the aperture size correction coefficient, the shutter speed correction coefficient, the sensitivity correction coefficient and the white balance correction coefficient. Using the image acquisition parameter correction parameters to correct the initial image acquisition parameters means multiplying the initial aperture size by the aperture size correction coefficient, multiplying the initial shutter speed by the shutter speed correction coefficient, multiplying the initial sensitivity by the sensitivity correction coefficient, and multiplying the initial white balance by the white balance correction coefficient, so as to obtain the dynamic image acquisition parameters.

[0064] In an embodiment of the present application, the dynamic point cloud acquisition parameters are determined based on the object properties of the inspection object, the initial point cloud acquisition parameters of the UAV's lidar, and the flight parameters of the UAV. The method for determining the dynamic point cloud acquisition parameters is described below.

[0065] In some embodiments, a controller of the drone obtains object attributes of the inspection object from a server. The controller determines point cloud acquisition parameter correction parameters based on the object attributes and the flight parameters. The controller corrects the initial point cloud acquisition parameters based on the point cloud acquisition parameter correction parameters to obtain the dynamic point cloud acquisition parameters.

[0066] Among them, the drone will transmit the initial point cloud collected in real time to the server. The initial point cloud collected in real time here is collected by the drone's laser radar based on the initial point cloud collection parameters. The server obtains the initial point cloud, performs target detection on the initial point cloud, and obtains the object properties of the inspection object in the initial point cloud. The server returns the object properties to the drone's controller, and the drone's controller obtains the object properties of the inspection object. The difference between the initial point cloud and the original point cloud is that the initial point cloud is collected based on the initial point cloud collection parameters, and the original point cloud is collected based on the dynamic point cloud collection parameters. The way the server identifies the object properties of the inspection object in the initial point cloud can adopt the target detection and target recognition methods in the relevant technology, or, after determining the object type of the inspection object, the server can use the object type to query the object properties of the inspection object in the object property database. This embodiment of the present application does not limit this. The point cloud acquisition parameter correction coefficient is used to correct the initial point cloud acquisition parameters to obtain a dynamic correction coefficient.

[0067] For example, the drone's controller obtains the object attributes of the inspection object from a server. The controller extracts features from the object attributes and the flight parameters to obtain object attribute features of the object attributes and flight parameter features of the flight parameters. The controller fuses the object attribute features and the flight parameter features to obtain a second correction parameter determination feature. The controller fully connects and normalizes the second correction parameter determination feature to obtain the point cloud acquisition parameter correction parameters. The controller uses the point cloud acquisition parameter correction parameters to correct the initial point cloud acquisition parameters to obtain the dynamic point cloud acquisition parameters.

[0068] Among them, the initial point cloud acquisition parameters include the initial field of view angle, the initial angular resolution, the initial sampling rate and the initial detection wavelength; the point cloud acquisition parameter correction parameters include the field of view angle correction coefficient, the angular resolution correction coefficient, the sampling rate correction coefficient and the detection wavelength correction coefficient; and using the point cloud acquisition parameter correction parameters to correct the initial point cloud acquisition parameters means multiplying the initial field of view angle by the field of view angle correction coefficient, multiplying the initial angular resolution by the angular resolution correction coefficient, multiplying the initial sampling rate by the sampling rate correction coefficient, and multiplying the initial detection wavelength by the detection wavelength correction coefficient, so as to obtain the dynamic point cloud acquisition parameters.

[0069] In some embodiments, the inspection instruction information includes an inspection target and an inspection strategy. The inspection instruction information is obtained by the server from a target terminal. The target terminal is a terminal bound to the drone, that is, a terminal used by the user.

[0070] 302. The server determines environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter.

[0071] The environmental image correction information is a parameter used to correct the original image, and the environmental point cloud correction information is a parameter used to correct the original point cloud. After correction, more accurate image and point cloud data can be obtained. The first environmental information includes first environmental light parameters, first environmental temperature, first environmental humidity, and weather type, and the second environmental information includes second environmental light parameters, second environmental temperature, and second environmental humidity.

[0072] In one possible implementation, the server determines an ambient light correction parameter based on the first ambient light parameter and the second ambient light parameter. The server determines a first temperature and humidity correction parameter and a second temperature and humidity correction parameter based on the first ambient temperature, the first ambient humidity, the second ambient temperature, the second ambient humidity, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter. The server determines a weather type correction parameter based on the weather type. The server determines the ambient image correction information and the ambient point cloud correction information based on the ambient light correction parameter, the first temperature and humidity correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter.

[0073] Those skilled in the art will appreciate that ambient light has a significant impact on image capture. The ambient light correction parameter is a parameter that corrects the image in combination with ambient light. Experiments have also revealed that ambient light also affects the collected point cloud, and changes in ambient light can result in varying degrees of noise in the collected point cloud. Those skilled in the art will appreciate that ambient temperature and humidity have a certain impact on image capture. In the embodiments of this application, the temperature and humidity correction parameter is determined in combination with ambient temperature, humidity, and dynamic image acquisition parameters. The dynamic image acquisition parameters are introduced to combine ambient temperature and humidity to amplify their impact on image capture. Those skilled in the art will appreciate that ambient humidity has a certain impact on point cloud capture. This is because water vapor in the atmosphere absorbs and scatters laser light, shortening the laser transmission distance and increasing signal attenuation, thereby affecting the detection range and accuracy of the LiDAR. Generally speaking, higher ambient humidity increases the impact. Furthermore, during experiments, it was discovered that temperature also has a certain impact on LiDAR point cloud capture. This impact may be due to differences in water vapor density. Therefore, temperature is also used to correct the point cloud in the embodiments of this application. Those skilled in the art can understand that weather type (sunny, rainy, snowy, hazy, foggy, etc.) has a huge impact on image capture and point cloud acquisition. Therefore, this application introduces weather type correction parameters based on ambient light correction parameters and temperature and humidity correction parameters to facilitate more comprehensive image correction and point cloud correction.

[0074] In order to explain the above embodiment more clearly, the above embodiment will be described in several parts below.

[0075] In the first part, the server determines an ambient light correction parameter based on the first ambient light parameter and the second ambient light parameter.

[0076] In one possible implementation, the first ambient light parameters include a first ambient light intensity, a first light color temperature, a first light direction, and a first light color; the second ambient light parameters include a second ambient light intensity, a second light color temperature, and a second light color; and the server determines ambient light intensity difference information based on the first ambient light intensity and the second ambient light intensity. The server determines light color temperature difference information based on the first light color temperature and the second light color temperature. The server determines intensity color temperature difference information based on the ambient light intensity difference information and the light color temperature difference information. The server determines light color difference information based on the first light direction, the first light color, and the second light color. The server determines the ambient light correction parameters based on the intensity color temperature difference information and the light color difference information.

[0077] For example, the server determines ambient light intensity difference information based on the first ambient light intensity and the second ambient light intensity. The server determines light color temperature difference information based on the first light color temperature and the second light color temperature. The server weightedly fuses the ambient light intensity difference information and the light color temperature difference information to obtain the intensity color temperature difference information. The server determines initial color difference information based on the first light color and the second light color. The server determines a directional color correction parameter based on the first light direction. The server multiplies the directional color correction parameter by the initial color difference information to obtain the light color difference information. The server splices the intensity color temperature difference information and the light color difference information to obtain ambient difference information. The server substitutes the ambient difference information into the first relationship data to obtain the ambient light correction parameter.

[0078] Among them, the weights of weighted fusion are set by technical personnel according to actual conditions, and the embodiments of the present application do not limit this. The first relationship data is a relationship function that can represent the relationship between environmental difference information and ambient light correction information. The first relationship data is set by technical personnel according to actual conditions, or is obtained by fitting a large number of parameters. The embodiments of the present application do not limit this. The correspondence between the first light direction and the directional color correction coefficient is set by technical personnel according to actual conditions. The server uses the first light direction to query in the first relationship table to obtain the corresponding directional color correction parameter. The first relationship table stores multiple first light directions and the directional color correction parameters corresponding to each first light direction. The first relationship table is set by technical personnel according to actual conditions.

[0079] In some embodiments, the ambient light correction parameter is a set of numbers, which will be subsequently converted into an ambient light correction matrix corresponding to the ambient light correction parameter. The size of the ambient light correction matrix is ​​the same as the pixel matrix of the original image. For example, if the pixel matrix of the original image is 2×2, and the ambient light correction parameter is (1, 2, 3, 4), then the ambient light correction parameter (1, 2, 3, 4) is converted into the ambient light correction matrix. .

[0080] In the second part, the server determines a first temperature and humidity correction parameter and a second temperature and humidity correction parameter based on the first ambient temperature, the first ambient humidity, the second ambient temperature, the second ambient humidity, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter.

[0081] In one possible implementation, the server determines an ambient temperature correction parameter based on the first ambient temperature and the second ambient temperature. The server determines an ambient humidity correction parameter based on the first ambient humidity and the second ambient humidity. The server determines the first temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter. The server determines the second temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter.

[0082] In order to explain the above embodiment more clearly, the above embodiment will be described in several parts below.

[0083] A. The server determines an ambient temperature correction parameter based on the first ambient temperature and the second ambient temperature.

[0084] In a possible implementation, the server determines ambient temperature difference information based on the first ambient temperature and the second ambient temperature, and determines the ambient temperature correction parameter based on the ambient temperature difference information.

[0085] For example, the server determines ambient temperature difference information based on the first ambient temperature and the second ambient temperature, and uses the ambient temperature difference information to query the second relationship table to obtain the ambient temperature correction parameter.

[0086] The second relationship table stores multiple candidate ambient temperature difference information and candidate ambient temperature correction parameters corresponding to each candidate ambient temperature difference information. The corresponding ambient temperature correction parameters can be obtained by querying the second relationship table using the ambient temperature difference information. The second relationship table is configured by a technician based on actual conditions and is not limited in this embodiment of the present application.

[0087] In some embodiments, the temperature correction parameter is a set of numbers, which will subsequently be transformed into a temperature correction matrix corresponding to the temperature correction parameter. The size of the temperature correction matrix is ​​the same as the pixel matrix of the original image. In addition, the size of the pixel matrix of the original image is the same as the size of the original point cloud.

[0088] B. The server determines an ambient humidity correction parameter based on the first ambient humidity and the second ambient humidity.

[0089] In a possible implementation, the server determines ambient humidity difference information based on the first ambient humidity and the second ambient humidity, and determines the ambient humidity correction parameter based on the ambient humidity difference information.

[0090] For example, the server determines ambient humidity difference information based on the first ambient humidity and the second ambient humidity, and uses the ambient humidity difference information to query the third relationship table to obtain the ambient humidity correction parameter.

[0091] The third relationship table stores multiple candidate ambient humidity difference information and candidate ambient humidity correction parameters corresponding to each candidate ambient humidity difference information. The corresponding ambient humidity correction parameters can be obtained by querying the third relationship table using the ambient humidity difference information. The third relationship table is configured by a technician based on actual conditions and is not limited in this embodiment of the present application.

[0092] In some embodiments, the humidity correction parameter is a set of numbers, which are subsequently transformed into a humidity correction matrix corresponding to the humidity correction parameter. The size of the humidity correction matrix is ​​the same as the pixel matrix of the original image.

[0093] C. The server determines the first temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter.

[0094] In one possible implementation, the server inputs the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter into a first correction parameter determination model, and uses the first correction parameter determination model to extract features of the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter to obtain a third correction parameter determination feature. The server performs multiple rounds of iterative decoding on the third correction parameter determination feature using the first correction parameter determination model to obtain the first temperature and humidity correction parameter.

[0095] Among them, the first correction parameter determination model includes a first encoder and a first decoder, the first encoder is used for feature extraction, and the first decoder is used for multiple rounds of iterative decoding. In some embodiments, the first encoder is an encoder constructed based on a fully connected network or a convolutional network, and the first decoder is a decoder constructed based on an attention mechanism. The first encoder and the first decoder can adopt encoders and decoders in related technologies. The embodiments of the present application do not limit the structure of the first encoder and the first decoder.

[0096] In some embodiments, the first temperature and humidity correction parameter is a set of numbers, which will be subsequently converted into a temperature and humidity correction matrix corresponding to the first temperature and humidity correction parameter. The size of the first temperature and humidity correction matrix is ​​the same as the pixel matrix of the original image.

[0097] D. The server determines the second temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter.

[0098] In one possible implementation, the server inputs the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter into a second correction parameter determination model. The server extracts features from the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter using the second correction parameter determination model to obtain a fourth correction parameter determination feature. The server performs multiple rounds of iterative decoding on the fourth correction parameter determination feature using the second correction parameter determination model to obtain the second temperature and humidity correction parameter.

[0099] Among them, the second correction parameter determination model includes a second encoder and a second decoder, the second encoder is used for feature extraction, and the second decoder is used for multiple rounds of iterative decoding. In some embodiments, the second encoder is an encoder constructed based on a fully connected network or a convolutional network, and the second decoder is a decoder constructed based on an attention mechanism. The second encoder and the second decoder can adopt the encoder and decoder in the related technology. The embodiment of the present application does not limit the structure of the second encoder and the second decoder.

[0100] In some embodiments, the second temperature and humidity correction parameter is a set of numbers, which will be subsequently transformed into a temperature and humidity correction matrix corresponding to the second temperature and humidity correction parameter. The size of the second temperature and humidity correction matrix is ​​the same as that of the original point cloud.

[0101] Part 3: The server determines the weather type correction parameter based on the weather type.

[0102] In a possible implementation, the server uses the weather type to query the fourth relationship table to obtain the weather type correction parameter.

[0103] Among them, the fourth relationship table stores multiple candidate weather types and candidate weather type correction parameters corresponding to each candidate weather type. The fourth relationship table is set by technical personnel according to actual conditions, and the embodiments of the present application do not limit this.

[0104] In some embodiments, the weather type correction parameter is a set of numbers, which will be subsequently transformed into a weather type correction matrix corresponding to the weather type correction parameter. The size of the weather type correction matrix is ​​the same as the pixel matrix and original point cloud of the original image.

[0105] Part 4: The server determines the environmental image correction information and the environmental point cloud correction information based on the environmental light correction parameter, the first temperature and humidity correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter.

[0106] In one possible implementation, the server performs a weighted fusion of the ambient light correction parameter, the first temperature and humidity correction parameter, and the weather type correction parameter to obtain the ambient image correction information. The server performs a weighted fusion of the ambient light correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter to obtain the ambient point cloud correction information.

[0107] For example, the server converts the ambient light correction parameter, the first temperature and humidity correction parameter, and the weather type correction parameter into an ambient light correction matrix, a first temperature and humidity correction matrix, and a weather type correction matrix, respectively. The server fuses the ambient light correction matrix, the first temperature and humidity correction matrix, and the weather type correction matrix to obtain an ambient image correction matrix, which serves as the ambient image correction information. The server converts the ambient light correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter into an ambient light correction matrix, a second temperature and humidity correction matrix, and a weather type correction matrix, respectively. The server fuses the ambient light correction matrix, the second temperature and humidity correction matrix, and the weather type correction matrix to obtain an ambient point cloud correction matrix, which serves as the ambient point cloud correction information.

[0108] Among them, the above-mentioned fusion of the ambient light correction matrix, the first temperature and humidity correction matrix, and the weather type correction matrix refers to weighted fusion of the ambient light correction matrix, the first temperature and humidity correction matrix, and the weather type correction matrix. Correspondingly, the above-mentioned fusion of the ambient light correction matrix, the second temperature and humidity correction matrix, and the weather type correction matrix refers to weighted fusion of the ambient light correction matrix, the second temperature and humidity correction matrix, and the weather type correction matrix. The weight of the weighted fusion is set by technical personnel according to actual conditions, and the embodiments of the present application do not limit this.

[0109] 303. The server determines image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection location, the environmental image correction information, and the environmental point cloud correction information.

[0110] Among them, the image enhancement parameters are parameters used to enhance the original image, and the point cloud enhancement parameters are parameters used to enhance the original point cloud. Both the image enhancement parameters and the point cloud enhancement parameters are enhancement parameters obtained by integrating multi-dimensional information.

[0111] In one possible implementation, the server determines image style description information based on the inspection target and inspection strategy in the inspection instruction information. The server determines point cloud noise description information based on the location attributes of the inspection location. The server determines the image enhancement parameters based on the image style description information and the environmental image correction information. The server determines the point cloud enhancement parameters based on the environmental point cloud correction information and the point cloud noise description information.

[0112] The image style description information and the point cloud noise description information are both in the form of natural language, which is convenient for technicians to review and perform subsequent processing. The position attributes of the inspection position are used to represent the characteristics of the inspection position. For example, the position attributes include position surface characteristics. The position surface characteristics are used to represent the flatness of the surface of the inspection position. The position surface characteristics will affect the reflection of the laser and will cause noise. Therefore, in the embodiments of the present application, they are used to determine the point cloud noise information. In some embodiments, the position surface characteristics are obtained by recognizing the original image.

[0113] In order to explain the above embodiment more clearly, the above embodiment will be described in several parts below.

[0114] In the first part, the server determines the image style description information based on the inspection target and inspection strategy in the inspection instruction information.

[0115] In one possible implementation, the server determines a baseline image style parameter corresponding to the inspection target based on the target description information of the inspection target. The server determines a policy adjustment parameter based on the inspection mode and inspection method in the inspection policy. The server determines the image style description information based on the baseline image style parameter and the policy adjustment parameter.

[0116] Among them, the image style description information includes a contrast enhancement coefficient, an edge sharpening level, and a color balance parameter. The target description information of the inspection target is obtained after the description method of the inspection target is converted. For example, the server encodes and decodes the inspection target based on the attention mechanism, and can convert the inspection target into the target description information of the inspection target. This is because the inspection target is configured by a technician and has various description methods. Converting it into target description information can achieve a unified description method and facilitate subsequent processing. The baseline image style parameter is used to describe the image style. The baseline image style parameter is similar to the filter in the related art. Inspection modes include single inspection and periodic inspection, etc. Single inspection includes single fast inspection and single fine inspection, etc. The requirements for the collected images under different inspection modes are different. The technicians will set it according to the requirements. The embodiments of this application do not limit this. Inspection modes include global inspection and local inspection, etc., which will be set by the technicians according to the requirements. The embodiments of this application do not limit this. In an embodiment of the present application, the style requirements for the collected images under different combinations of inspection modes and inspection methods are differentiated to adapt to the needs of different inspection modes and inspection methods. The strategy adjustment parameters determined based on the inspection modes and inspection methods are used to meet the image style requirements of different inspection modes and inspection methods.

[0117] For example, the server determines multiple style keywords that match the target description information. These multiple style keywords form a style keyword set, which includes at least two of contrast level, saturation, sharpness intensity, and hue tendency. The server performs semantic encoding on the multiple style keywords to obtain keyword semantic features for each style keyword. The server fully connects and normalizes the keyword semantic features of the multiple style keywords to obtain visual feature parameters. The server uses the visual feature parameters to query a style mapping table to obtain the baseline image style parameters. The style mapping table is used to represent the correspondence between visual feature parameters and baseline image style parameters. The server uses the inspection mode and the inspection method to query a fifth relationship table to obtain the policy adjustment parameter. The server performs feature extraction on the baseline image style parameter and the policy adjustment parameter, respectively, to obtain style parameter features of the baseline image style parameter and adjustment parameter features of the policy adjustment parameter. The server fuses the style parameter features and the adjustment parameter features to obtain style description features. The server performs multiple rounds of iterative decoding on the style description features to obtain the image style description information.

[0118] Among them, the fifth relationship table stores multiple candidate inspection modes, multiple candidate inspection methods, and multiple candidate policy adjustment parameters. One candidate policy adjustment parameter corresponds to one candidate inspection mode and one candidate inspection method. The fifth relationship table is set by technicians according to actual conditions and is not limited in this embodiment of the present application. Multi-round iterative decoding is multi-round iterative decoding based on the attention mechanism.

[0119] In the second part, the server determines the point cloud noise description information based on the location attributes of the inspection location.

[0120] In one possible implementation, the server extracts features from the surface characteristics of the location in the location attribute to obtain the surface features of the inspection location. The server performs multiple rounds of iterative decoding on the surface features based on an attention mechanism to obtain the point cloud noise description information.

[0121] Part three: The server determines the image enhancement parameters based on the image style description information and the environmental image correction information.

[0122] In one possible implementation, the server generates dynamic contrast control parameters based on the contrast enhancement coefficient in the image style description information and the ambient light correction parameter in the environmental image correction information. The server optimizes the edge sharpening level in the image style description information based on the first temperature and humidity correction parameter in the environmental image correction information to generate a sharpening intensity parameter that is resistant to environmental interference. The server generates environmentally adaptive color correction parameters based on the weather type correction parameter in the environmental image correction information and the color balance parameter in the image style description information.

[0123] The image enhancement parameters include the dynamic contrast control parameter, the sharpening intensity parameter for resisting environmental interference, and the color correction parameter for environmental adaptation.

[0124] For example, the server combines the contrast enhancement coefficient in the image style description information with the ambient light correction parameter in the environmental image correction information to obtain the dynamic contrast control parameter. The server determines the edge sharpening level correction parameter based on the first temperature and humidity correction parameter in the environmental image correction information. The server uses the edge sharpening level correction parameter to correct the edge sharpening level in the image style description information to obtain a sharpening intensity parameter that is resistant to environmental interference. The server performs feature extraction on the weather type correction parameter in the environmental image correction information and the color balance parameter in the image style description information to obtain a color correction feature. The server performs full connection and normalization on the color correction feature to obtain the color correction parameter.

[0125] The correspondence between the first temperature and humidity correction parameter and the edge sharpening level correction parameter is set by a technician based on actual conditions and is not limited in this embodiment of the present application. Using the edge sharpening level correction parameter to correct the edge sharpening level in the image style description information refers to multiplying the edge sharpening level correction parameter by the edge sharpening level and then adding the result to the edge sharpening level.

[0126] Part 4: The server determines the point cloud enhancement parameters based on the environmental point cloud correction information and the point cloud noise description information.

[0127] In one possible implementation, the server performs feature extraction on the environmental point cloud correction information and the point cloud noise description information to obtain environmental point cloud correction features of the environmental point cloud correction information and point cloud noise description features of the point cloud noise description information. The server fuses the environmental point cloud correction features and the point cloud noise description features to obtain point cloud enhancement features. The server performs multiple rounds of iterative decoding on the point cloud enhancement features to obtain point cloud enhancement parameters.

[0128] Among them, the point cloud enhancement parameters include dynamic noise suppression parameters and point cloud density compensation parameters.

[0129] It should be noted that, in the embodiments of the present application, the encoding based on the attention mechanism and the decoding based on the attention mechanism are used many times. The process of encoding based on the attention mechanism can be implemented by an encoder, which can be an encoder based on the attention mechanism in the relevant technology, such as an encoder for the BERT network. Correspondingly, the process of decoding based on the attention mechanism can be implemented by a decoder, which can be a decoder based on the attention mechanism in the relevant technology, such as a decoder for the BERT network. Of course, in addition to BERT, it can also be an encoder and decoder in other models based on the attention mechanism. The embodiments of the present application do not limit this. In addition, the full connection in the embodiments of the present application refers to multiplying the parameters or features with the full connection matrix, and normalization refers to substituting the parameters or features into the normalization function and processing them by the normalization function. The full connection matrix is ​​obtained through training, and the normalization function is selected by the technician according to the actual situation, such as selecting the SoftMax function or the ReLu function as the normalization function, etc. The embodiments of the present application do not limit this.

[0130] 304. The server processes the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters to obtain a target image and a target point cloud.

[0131] Among them, the target image and target point cloud are obtained after enhancement. Compared with the original image and original point cloud, they have more information and higher accuracy.

[0132] In one possible implementation, the server performs contrast enhancement, edge sharpening, and color correction on the original image based on the dynamic contrast control parameters, environmental interference resistance sharpening intensity parameters, and environmental adaptive color correction parameters in the image enhancement parameters to generate an intermediate optimized image. Based on the dynamic noise suppression parameters and point cloud density compensation parameters in the point cloud enhancement parameters, the server performs denoising filtering and density compensation on the original point cloud to generate an intermediate optimized point cloud. The server then registers and aligns the intermediate optimized image with the intermediate optimized point cloud to obtain a target image and target point cloud that are temporally and spatially consistent.

[0133] Among them, the point cloud enhancement parameters include dynamic noise suppression parameters generated based on the environmental point cloud correction information and density compensation parameters generated based on the position surface characteristics of the inspection position. The process of generating dynamic noise suppression parameters based on the environmental point cloud correction information is to perform feature extraction, full connection and normalization on the environmental point cloud correction information, and the process of generating density compensation parameters based on the position surface characteristics is to perform feature extraction, full connection and normalization on the position surface characteristics. The purpose of registration is to find the correspondence between the pixel points in the image and the data points in the point cloud, so as to complete the registration of the image and the point cloud. The specific process involves technical methods such as coordinate transformation and similarity matching between pixel points and data points in the point cloud. The specific implementation process can refer to the processing method in the relevant technology, and the embodiments of this application are not limited to this.

[0134] 305. The server determines whether there is any abnormality at the inspection location based on the target image and the target point cloud.

[0135] In one possible method, the server performs multi-scale feature extraction on the target image to generate image texture features and shape features. The server performs spatial distribution analysis on the target point cloud to extract point cloud density features and geometric topology features. The server performs multimodal feature fusion on the image texture features and shape features with the point cloud density features and geometric topology features to obtain abnormal features. The server calculates an abnormality score for the abnormal feature using an abnormality judgment model. When the score exceeds a preset judgment threshold, it is determined that an abnormality exists at the inspection location. When it is determined that an abnormality exists, the server outputs the abnormality type with the highest matching degree in the abnormality type set and the three-dimensional spatial coordinates of the inspection location. The abnormality type set is dynamically adapted to the inspection indication information. When the score does not exceed the preset judgment threshold, it is determined that no abnormality exists at the inspection location.

[0136] Among them, the abnormality judgment model is obtained through multiple rounds of training with multiple sample data and the annotation information of each sample data. One sample data includes a sample image and a sample point cloud. The annotation information is used to indicate whether there is an abnormality and the type of abnormality at the inspection location corresponding to the sample data. In the process of training the abnormality judgment model, a comparative learning method can be adopted, that is, the multiple sample data are divided into positive sample data and negative sample data. The annotation information corresponding to the positive sample data indicates that there is no abnormality, and the annotation information corresponding to the negative sample data indicates that there is an abnormality. The abnormality type set includes multiple abnormality types. The corresponding abnormality type set can be found through the inspection indication information, that is, there are multiple candidate abnormality type sets. After obtaining the inspection indication information, the abnormality type set that matches the inspection indication information can be determined from the multiple candidate abnormality type sets. The preset judgment threshold is set by technical personnel according to actual conditions, and the embodiments of the present application do not limit this. In addition,

[0137] The sample images and sample point clouds used in training the anomaly determination model are the target sample images and target sample point clouds obtained after the original sample images and original point clouds are processed by the above steps 302-304. The original sample images are images collected by the drone, and the original sample point clouds are point clouds collected by the drone's lidar. During the experiment, the accuracy of anomaly detection directly based on the original sample images and sample point clouds is much lower than the result of anomaly detection based on the target sample images and target sample point clouds. This is because the target sample images and target sample point clouds can carry richer information after being processed by the above steps 302-303. The above training actually refers to a fine-tuning process, that is, in the embodiment of the present application, a pre-trained multimodal model can be used, and then the multimodal model can be supervised and fine-tuned using sample images, sample point clouds and corresponding annotation information to finally obtain the anomaly determination model. The specific method of supervised fine-tuning has not been improved in the technical solution provided in the embodiment of the present application, and reference can be made to the methods in the relevant technology.

[0138] In order to explain the above embodiment more clearly, the process of calculating the abnormality score of the abnormal feature by the server through the abnormality determination model in the above embodiment is explained below.

[0139] In some embodiments, the server inputs the abnormality feature into the abnormality determination model, and performs full concatenation and normalization on the abnormality feature through the score mapping unit of the abnormality determination model to obtain a score value. The server performs full concatenation and normalization on the abnormality feature through the abnormality type classification unit of the abnormality recognition model to obtain a probability set for the abnormality type set, where the probability set includes multiple probabilities, with each probability corresponding to an abnormality type in the abnormality type set.

[0140] The anomaly type with the highest matching degree in the anomaly type set refers to the anomaly type that matches the inspection location, that is, the anomaly type with the highest probability in the probability set.

[0141] 306. When determining whether there is an abnormality at the inspection location, the server sends abnormality prompt information to the target terminal. The abnormality prompt information is used to prompt that there is an abnormality at the inspection location.

[0142] The abnormal prompt information carries the abnormal type of the inspection location and the three-dimensional space coordinates of the inspection location.

[0143] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0144] Through the technical solution provided in the embodiment of the present application, the original image collected by the image acquisition component of the UAV, the original point cloud collected by the lidar, the first environmental information of the environment in which the UAV is located, the second environmental information of the inspection position of the UAV and the dynamic acquisition parameters are obtained. Based on the first environmental information, the second environmental information, the dynamic image acquisition parameters and the dynamic point cloud acquisition parameters, the environmental image correction information and the environmental point cloud correction information are determined. Based on the inspection indication information, the inspection position, the environmental image correction information and the environmental point cloud correction information, the image enhancement parameters and the point cloud enhancement parameters are determined. Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain the target image and the target point cloud. Based on the target image and the target point cloud, it is determined whether there is an abnormality at the inspection position. Multi-dimensional information is combined during the inspection, the inspection effect is higher, and anomalies in a specific space can be discovered in a timely manner.

[0145] Figure 4 This is a schematic diagram of a large-space UAV automatic inspection system based on laser radar provided in an embodiment of the present application. Figure 4 The system includes: an acquisition module 401, a correction information determination module 402, an enhancement parameter determination module 403, a processing module 404 and an abnormality identification module 405.

[0146] The acquisition module 401 is used to obtain the original image captured by the image acquisition component of the drone, the original point cloud captured by the lidar, the first environmental information of the drone's environment, the second environmental information of the drone's inspection position, and dynamic acquisition parameters, which include inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters.

[0147] The correction information determination module 402 is configured to determine the environmental image correction information and the environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters.

[0148] The enhancement parameter determination module 403 is configured to determine image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information.

[0149] The processing module 404 is configured to process the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters, respectively, to obtain a target image and a target point cloud.

[0150] The abnormality identification module 405 is used to determine whether there is an abnormality in the inspection position based on the target image and the target point cloud.

[0151] In one possible implementation, the first environmental information includes a first environmental light parameter, a first environmental temperature, a first environmental humidity, and a weather type; the second environmental information includes a second environmental light parameter, a second environmental temperature, and a second environmental humidity; and the correction information determination module 402 is configured to determine an environmental light correction parameter based on the first environmental light parameter and the second environmental light parameter. Determine a first temperature and humidity correction parameter and a second temperature and humidity correction parameter based on the first environmental temperature, the first environmental humidity, the second environmental temperature, the second environmental humidity, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters. Determine a weather type correction parameter based on the weather type. Determine the environmental image correction information and the environmental point cloud correction information based on the environmental light correction parameter, the first temperature and humidity correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter.

[0152] In one possible implementation, the first ambient light parameters include a first ambient light intensity, a first light color temperature, a first light direction, and a first light color; and the second ambient light parameters include a second ambient light intensity, a second light color temperature, and a second light color. The correction information determination module 402 is configured to determine ambient light intensity difference information based on the first ambient light intensity and the second ambient light intensity; determine light color temperature difference information based on the first light color temperature and the second light color temperature; determine intensity color temperature difference information based on the ambient light intensity difference information and the light color temperature difference information; determine light color difference information based on the first light direction, the first light color, and the second light color; and determine the ambient light correction parameters based on the intensity color temperature difference information and the light color difference information.

[0153] In one possible implementation, the correction information determination module 402 is configured to determine an ambient temperature correction parameter based on the first ambient temperature and the second ambient temperature; determine an ambient humidity correction parameter based on the first ambient humidity and the second ambient humidity; determine the first temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter; and determine the second temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter.

[0154] In one possible implementation, the enhancement parameter determination module 403 is configured to determine image style description information based on the inspection target and inspection strategy in the inspection instruction information; determine point cloud noise description information based on the location attributes of the inspection location; determine image enhancement parameters based on the image style description information and the environmental image correction information; and determine point cloud enhancement parameters based on the environmental point cloud correction information and the point cloud noise description information.

[0155] In one possible implementation, the enhancement parameter determination module 403 is configured to determine baseline image style parameters corresponding to the inspection target based on the target description information of the inspection target. The strategy adjustment parameters are determined based on the inspection mode and inspection method in the inspection strategy. The image style description information is determined based on the baseline image style parameters and the strategy adjustment parameters. The image style description information includes a contrast enhancement coefficient, an edge sharpening level, and a color balance parameter.

[0156] In one possible embodiment, the enhancement parameter determination module 403 is used to generate a dynamic contrast control parameter based on the contrast enhancement coefficient in the image style description information and the ambient light correction parameter in the environmental image correction information. The edge sharpening level in the image style description information is parameter-optimized based on the first temperature and humidity correction parameter in the environmental image correction information to generate a sharpening intensity parameter for resisting environmental interference. The environment-adaptive color correction parameter is generated based on the weather type correction parameter in the environmental image correction information and the color balance parameter in the image style description information. The image enhancement parameters include the dynamic contrast control parameter, the sharpening intensity parameter for resisting environmental interference, and the environment-adaptive color correction parameter.

[0157] In one possible embodiment, the processing module 404 is used to perform contrast enhancement, edge sharpening and color correction processing on the original image based on the dynamic contrast control parameters, the sharpening intensity parameters for resisting environmental interference and the color correction parameters for environmental adaptation in the image enhancement parameters to generate an intermediate optimized image. Based on the dynamic noise suppression parameters and the point cloud density compensation parameters in the point cloud enhancement parameters, the original point cloud is subjected to denoising filtering and density compensation processing to generate an intermediate optimized point cloud. The intermediate optimized image and the intermediate optimized point cloud are aligned to obtain a target image and target point cloud with temporal and spatial consistency. The point cloud enhancement parameters include dynamic noise suppression parameters generated based on the environmental point cloud correction information and density compensation parameters generated based on the positional surface characteristics of the inspection position.

[0158] In one possible embodiment, the anomaly identification module 405 is used to perform multi-scale feature extraction on the target image to generate image texture features and shape features. The spatial distribution analysis of the target point cloud is performed to extract point cloud density features and geometric topological features. The image texture features and shape features are fused with the point cloud density features and geometric topological features to obtain anomaly features. The anomaly judgment model is used to calculate an anomaly score for the anomaly feature. When the score exceeds a preset judgment threshold, it is determined that an anomaly exists at the inspection location. When it is determined that an anomaly exists, the anomaly type with the highest matching degree in the anomaly type set and the three-dimensional spatial coordinates of the inspection location are output, and the anomaly type set is dynamically adapted to the inspection indication information.

[0159] It should be noted that the above-mentioned embodiment provides a large-space drone automatic inspection system based on laser radar. The above-mentioned functional modules are only used as examples to illustrate the division of the above-mentioned functional modules during inspection. In actual applications, the above-mentioned functions can be distributed to different functional modules as needed, that is, the internal structure of the computer equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the large-space drone automatic inspection system based on laser radar and the large-space drone automatic inspection method based on laser radar provided in the above-mentioned embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0160] Through the technical solution provided in the embodiment of the present application, the original image collected by the image acquisition component of the UAV, the original point cloud collected by the lidar, the first environmental information of the environment in which the UAV is located, the second environmental information of the inspection position of the UAV and the dynamic acquisition parameters are obtained. Based on the first environmental information, the second environmental information, the dynamic image acquisition parameters and the dynamic point cloud acquisition parameters, the environmental image correction information and the environmental point cloud correction information are determined. Based on the inspection indication information, the inspection position, the environmental image correction information and the environmental point cloud correction information, the image enhancement parameters and the point cloud enhancement parameters are determined. Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain the target image and the target point cloud. Based on the target image and the target point cloud, it is determined whether there is an abnormality at the inspection position. Multi-dimensional information is combined during the inspection, the inspection effect is higher, and anomalies in a specific space can be discovered in a timely manner.

[0161] Figure 5 This is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 500 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 501 and one or more memories 502, wherein the one or more memories 502 store at least one computer program, and the at least one computer program is loaded and executed by the one or more processors 501 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server 500 may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The server 500 may also include other components for implementing device functions, which will not be described in detail here.

[0162] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory device including a computer program. The computer program can be executed by a processor to implement the lidar-based large-space drone automatic inspection method described in the above embodiment. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, or an optical data storage device.

[0163] In an exemplary embodiment, a computer program product or computer program is also provided, which includes a program code, which is stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium, and the processor executes the program code, so that the computer device executes the above-mentioned large-space drone automatic inspection method based on lidar.

[0164] In some embodiments, the computer program involved in the embodiments of the present application may be deployed and executed on a computer device, or on multiple computer devices located at one location, or on multiple computer devices distributed at multiple locations and interconnected through a communication network. Multiple computer devices distributed at multiple locations and interconnected through a communication network may constitute a blockchain system.

[0165] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0166] The above are only optional embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A large-space UAV automatic inspection method based on laser radar, characterized in that: The method comprises: Obtaining an original image captured by an image acquisition component of a drone, an original point cloud captured by a lidar, first environmental information of an environment in which the drone is located, second environmental information of an inspection location of the drone, and dynamic acquisition parameters, the dynamic acquisition parameters including inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters; Determining environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters; Determining image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information; Based on the image enhancement parameters and the point cloud enhancement parameters, the original image and the original point cloud are processed respectively to obtain a target image and a target point cloud; The processing of the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters to obtain a target image and a target point cloud includes: Based on the dynamic contrast control parameters, the sharpening intensity parameters for resisting environmental interference and the color correction parameters for environmental adaptation in the image enhancement parameters, the original image is subjected to contrast enhancement, edge sharpening and color correction processing to generate an intermediate optimized image; based on the dynamic noise suppression parameters and the point cloud density compensation parameters in the point cloud enhancement parameters, the original point cloud is subjected to denoising filtering and density compensation processing to generate an intermediate optimized point cloud; the intermediate optimized image and the intermediate optimized point cloud are registered and aligned to obtain a target image and target point cloud with temporal and spatial consistency; wherein the point cloud enhancement parameters include dynamic noise suppression parameters generated based on the environmental point cloud correction information and density compensation parameters generated based on the positional surface characteristics of the inspection position; Based on the target image and the target point cloud, it is determined whether there is any abnormality in the inspection position.

2. The method according to claim 1, characterized in that The first environmental information includes a first environmental light parameter, a first environmental temperature, a first environmental humidity, and a weather type; the second environmental information includes a second environmental light parameter, a second environmental temperature, and a second environmental humidity; and determining the environmental image correction information and the environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter includes: determining an ambient light correction parameter based on the first ambient light parameter and the second ambient light parameter; Determining a first temperature and humidity correction parameter and a second temperature and humidity correction parameter based on the first ambient temperature, the first ambient humidity, the second ambient temperature, the second ambient humidity, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter; determining a weather type correction parameter based on the weather type; The environmental image correction information and the environmental point cloud correction information are determined based on the environmental light correction parameter, the first temperature and humidity correction parameter, the second temperature and humidity correction parameter, and the weather type correction parameter.

3. The method according to claim 2, characterized in that The first ambient light parameters include a first ambient light intensity, a first light color temperature, a first light direction, and a first light color; the second ambient light parameters include a second ambient light intensity, a second light color temperature, and a second light color; and determining the ambient light correction parameter based on the first ambient light parameters and the second ambient light parameters includes: determining ambient light intensity difference information based on the first ambient light intensity and the second ambient light intensity; determining light color temperature difference information based on the first light color temperature and the second light color temperature; Determining intensity and color temperature difference information based on the ambient light intensity difference information and the light color temperature difference information; determining light color difference information based on the first light direction, the first light color, and the second light color; The ambient light correction parameter is determined based on the intensity-color temperature difference information and the light color difference information.

4. The method according to claim 2, characterized in that The determining of the first temperature and humidity correction parameter and the second temperature and humidity correction parameter based on the first ambient temperature, the first ambient humidity, the second ambient temperature, the second ambient humidity, the dynamic image acquisition parameter, and the dynamic point cloud acquisition parameter includes: determining an ambient temperature correction parameter based on the first ambient temperature and the second ambient temperature; determining an ambient humidity correction parameter based on the first ambient humidity and the second ambient humidity; Determining the first temperature and humidity correction parameter based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic image acquisition parameter; The second temperature and humidity correction parameter is determined based on the ambient temperature correction parameter, the ambient humidity correction parameter, and the dynamic point cloud acquisition parameter.

5. The method according to claim 1, wherein The determining of image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information includes: Determining image style description information based on the inspection target and inspection strategy in the inspection instruction information; Determining point cloud noise description information based on the location attributes of the inspection location; Determining the image enhancement parameters based on the image style description information and the environmental image correction information; The point cloud enhancement parameters are determined based on the environmental point cloud correction information and the point cloud noise description information.

6. The method according to claim 5, characterized in that The determining of the image style description information based on the inspection target and the inspection strategy in the inspection instruction information includes: Determining a reference image style parameter corresponding to the inspection target based on the target description information of the inspection target; Determining policy adjustment parameters based on the inspection mode and inspection method in the inspection policy; Determining the image style description information based on the baseline image style parameter and the strategy adjustment parameter; The image style description information includes contrast enhancement coefficient, edge sharpening level and color balance parameter.

7. The method according to claim 5, characterized in that The determining the image enhancement parameters based on the image style description information and the environmental image correction information includes: generating a dynamic contrast control parameter based on the contrast enhancement coefficient in the image style description information and the ambient light correction parameter in the ambient image correction information; Optimizing the edge sharpening level in the image style description information based on the first temperature and humidity correction parameter in the environmental image correction information to generate a sharpening intensity parameter that is resistant to environmental interference; generating environment-adaptive color correction parameters based on the weather type correction parameters in the environment image correction information and the color balance parameters in the image style description information; The image enhancement parameters include the dynamic contrast control parameters, the sharpening intensity parameters for resisting environmental interference, and the color correction parameters for environmental adaptation.

8. The method according to claim 1, characterized in that The determining whether there is an abnormality at the inspection location based on the target image and the target point cloud includes: Performing multi-scale feature extraction on the target image to generate image texture features and shape features; Performing spatial distribution analysis on the target point cloud to extract point cloud density features and geometric topological features; Performing multimodal feature fusion on the image texture features, shape features, point cloud density features, and geometric topology features to obtain abnormal features; An abnormality score is calculated for the abnormal feature through an abnormality judgment model, and when the score value exceeds a preset judgment threshold, it is determined that an abnormality exists at the inspection location; When it is determined that an abnormality exists, the abnormality type with the highest matching degree in the abnormality type set and the three-dimensional spatial coordinates of the inspection position are output, and the abnormality type set is dynamically adapted to the inspection indication information.

9. A large-space UAV automatic inspection system based on laser radar, characterized in that: The system comprises: An acquisition module is configured to acquire an original image captured by the image acquisition component of the drone, an original point cloud captured by the lidar, first environmental information of the drone's environment, second environmental information of the drone's inspection location, and dynamic acquisition parameters, wherein the dynamic acquisition parameters include inspection indication information, dynamic image acquisition parameters, and dynamic point cloud acquisition parameters; a correction information determination module, configured to determine environmental image correction information and environmental point cloud correction information based on the first environmental information, the second environmental information, the dynamic image acquisition parameters, and the dynamic point cloud acquisition parameters; an enhancement parameter determination module, configured to determine image enhancement parameters and point cloud enhancement parameters based on the inspection instruction information, the inspection position, the environmental image correction information, and the environmental point cloud correction information; a processing module, configured to process the original image and the original point cloud based on the image enhancement parameters and the point cloud enhancement parameters, respectively, to obtain a target image and a target point cloud; The processing module is used to perform contrast enhancement, edge sharpening and color correction processing on the original image based on the dynamic contrast control parameters, anti-environmental interference sharpening intensity parameters and environmental adaptive color correction parameters in the image enhancement parameters to generate an intermediate optimized image; perform denoising filtering and density compensation processing on the original point cloud based on the dynamic noise suppression parameters and point cloud density compensation parameters in the point cloud enhancement parameters to generate an intermediate optimized point cloud; and perform registration and alignment on the intermediate optimized image and the intermediate optimized point cloud to obtain a target image and target point cloud with spatiotemporal consistency; wherein the point cloud enhancement parameters include dynamic noise suppression parameters generated based on the environmental point cloud correction information and density compensation parameters generated based on the position surface characteristics of the inspection position; The abnormality recognition module is used to determine whether there is an abnormality in the inspection position based on the target image and the target point cloud.

Images