Unmanned aerial vehicle emergency scene inspection method and system based on deep reinforcement learning

Abstract

The invention relates to the technical field of unmanned aerial vehicle inspection, in particular to an unmanned aerial vehicle emergency scene inspection method and system based on deep reinforcement learning, and a data processing module is used for preprocessing meteorological data to obtain optimized meteorological data; the model training module is used for processing the image data and constructing an emergency scene model; the path planning module is used for performing optimal path planning of the unmanned aerial vehicle based on the optimized meteorological data and the constructed emergency scene model; therefore, the inspection path of the unmanned aerial vehicle is optimized based on the meteorological condition and the three-dimensional model, so that the unmanned aerial vehicle can avoid inspection accidents caused by meteorology, the unmanned aerial vehicle can fly in the safest and most efficient area of an emergency scene, inspection can be performed in the optimal path all the time, manual takeover is not needed, and the labor intensity of workers is reduced. And the inspection efficiency is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV inspection, and particularly to a UAV emergency scenario inspection method and system based on deep reinforcement learning. Background Art

[0002] Currently, in natural disasters, due to the damage caused by earthquakes, floods or tsunamis, local communication facilities are usually damaged, making it impossible for rescue workers to communicate with the local area in real time. Therefore, when a natural disaster occurs, in order to ensure smooth communication, UAVs need to take off emergently to conduct inspections in emergency scenarios and establish remote communication with the fire control background so that rescue workers can fully understand the on-site situation.

[0003] In the aforementioned prior art, when a UAV conducts inspections, it usually relies on manual operation for flight. However, the area affected by natural disasters is usually large and the flight time is long, which is likely to cause fatigue among the staff. Therefore, an algorithm is used to set the route to replace manual operation. However, relying on the method of setting the route, the UAV can only fly according to a fixed route. Once there are situations that affect flight and communication in the route, the UAV cannot avoid them autonomously, thereby affecting the inspection efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide a UAV emergency scenario inspection method and system based on deep reinforcement learning to solve the problems in the prior art.

[0005] To achieve the above purpose, the present invention provides a UAV emergency scenario inspection system based on deep reinforcement learning, including an emergency inspection communication module, an environment perception and acquisition module, a data processing module, a model training module, and a path planning module. The emergency inspection communication module, the environment perception and acquisition module, the data processing module, the model training module, and the path planning module are connected in sequence;

[0006] The emergency inspection communication module is used to set inspection communication equipment on the UAV to establish a real-time communication connection with the host computer during emergency scenario inspections;

[0007] The environment perception and acquisition module is used to set sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data;

[0008] The data processing module is used to preprocess the meteorological data to obtain optimized meteorological data;

[0009] The model training module is used to process the image data to construct an emergency scenario model;

[0010] The path planning module is used to perform optimal path planning for the UAV based on the optimized meteorological data and the constructed emergency scenario model.

[0011] Among them, the environmental perception and acquisition module includes a terrain pre-scanning unit, a light acquisition unit, a wind direction and speed acquisition unit, a weather acquisition unit, and a meteorological data integration unit. The terrain pre-scanning unit, the light acquisition unit, the wind direction and speed acquisition unit, the weather acquisition unit, and the meteorological data integration unit are connected in sequence;

[0012] The terrain pre-scanning unit is used to scan the terrain of the emergency scene before emergency inspection to obtain image data for constructing an emergency scene model;

[0013] The light acquisition unit is used to collect the light intensity and light angle during the emergency scene inspection to obtain light data;

[0014] The wind direction and speed acquisition unit is used to collect the wind direction and speed during the emergency scene inspection to obtain wind direction and speed data;

[0015] The weather acquisition unit is used to collect whether it is raining, whether there is lightning, and the temperature and humidity during the emergency scene inspection to obtain weather data;

[0016] The meteorological data integration unit is used to integrate the light data, wind direction and speed data, and weather data to obtain meteorological data.

[0017] Among them, the model training module includes an emergency area terrain construction unit and a data visualization display unit. The emergency area terrain construction unit and the data visualization display unit are connected;

[0018] The emergency area terrain construction unit is used to construct the emergency scene model according to the image data;

[0019] The data visualization display unit is used to map the meteorological data onto the emergency scene model.

[0020] Among them, the emergency area terrain construction unit includes a graphics preprocessing subunit, a 3D reconstruction subunit, a texture mapping subunit, and a reconstruction optimization subunit. The graphics preprocessing subunit, the 3D reconstruction subunit, the texture mapping subunit, and the reconstruction optimization subunit are connected in sequence;

[0021] The graphics preprocessing subunit is used to preprocess the image data to obtain preprocessed image data;

[0022] The 3D reconstruction subunit is used to extract point clouds from the preprocessed image data and construct a three-dimensional grid model of the emergency scene based on the point cloud data using the Poisson reconstruction algorithm;

[0023] The texture mapping subunit is used to project the original image onto the surface of the three-dimensional grid model of the emergency scene, and then optimize the texture seams through a Markov random field to obtain the emergency scene model;

[0024] The reconstruction and optimization subunit is used to perform smoothing processing and hole filling on the emergency scene model.

[0025] Among them, the data visualization and display unit includes a time synchronization subunit, a space synchronization subunit, a data normalization subunit, and a display scaling subunit, and the time synchronization subunit, the space synchronization subunit, the data normalization subunit, and the display scaling subunit are connected in sequence;

[0026] The time synchronization subunit is used to synchronize the information of the optimized meteorological data with the time of the emergency scene model to reduce errors;

[0027] The space synchronization subunit is used to synchronize the geographical coordinates of the optimized meteorological data onto the emergency scene model;

[0028] The data normalization subunit is used to normalize the illumination, wind direction and speed, temperature and humidity information of the optimized meteorological data to the interval [0, 1] and display it on the emergency scene model;

[0029] The display scaling subunit is used to add a scaling function to the displayed optimized meteorological data, and the display size can be scaled.

[0030] Among them, the path planning module includes a path preset unit, a meteorological real-time update unit, a meteorological threshold trigger unit, a patrol route switching unit, a UAV current state reading unit, a UAV action control unit, and an emergency adjustment unit, and the path preset unit, the meteorological real-time update unit, the meteorological threshold trigger unit, the patrol route switching unit, the UAV current state reading unit, the UAV action control unit, and the emergency adjustment unit are connected in sequence;

[0031] The path preset unit is used to, when there is no emergency patrol, the staff pre-sets the most efficient flight path and multiple alternative flight paths in the UAV in advance, and if there is no meteorological anomaly, it flies according to the preset path;

[0032] The meteorological real-time update unit is used to update the optimized meteorological data on the emergency scene model in real time;

[0033] The meteorological threshold trigger unit is used to set corresponding first-level thresholds for each item of the optimized meteorological data respectively. When any data exceeds the corresponding first-level threshold, it indicates that there is a situation affecting the safety or flight efficiency of the UAV. At this time, the patrol route switching unit is started to switch the patrol route;

[0034] The inspection route switching unit is used to automatically switch the standby flight path of the UAV after any data exceeds the first-level threshold. If the first-level threshold is still triggered after multiple flight path switches, the UAV action control unit performs path micro-adjustment;

[0035] The UAV current state reading unit is used to read the current flight speed, direction and power of the UAV for the staff to observe at any time;

[0036] The UAV action control unit is used to identify the specific area of the data on the emergency scenario model and control the UAV to fly out of the meteorological coverage range of the data when the meteorological threshold is triggered again after switching flight, and bypass this area in subsequent inspections;

[0037] The emergency adjustment unit is used to set the second-level threshold. When the first-level threshold exists in all areas of the emergency scenario model but does not exceed the second-level threshold, it controls the UAV to avoid the meteorological coverage area of the second-level threshold. When all areas exceed the second-level threshold, it controls the UAV to return.

[0038] The present invention also provides a UAV emergency scenario inspection method based on deep reinforcement learning, which adopts the above-mentioned UAV emergency scenario inspection system based on deep reinforcement learning, and includes the following steps:

[0039] Set inspection communication equipment on the UAV to establish a real-time communication connection with the upper computer during emergency scenario inspections;

[0040] Set sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data;

[0041] Preprocess the meteorological data to obtain optimized meteorological data;

[0042] Process the image data to construct an emergency scenario model;

[0043] Based on the optimized meteorological data and the constructed emergency scenario model, perform optimal path planning for the UAV;

[0044] The UAV starts to inspect the emergency scenario.

[0045] An unmanned aerial vehicle (UAV) emergency scenario inspection method and system based on deep reinforcement learning. The emergency inspection communication module is used to set up inspection communication equipment on the UAV to establish a real-time communication connection with the host computer during emergency scenario inspections. The environmental perception and acquisition module is used to set up sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data. The data processing module is used to preprocess the meteorological data to obtain optimized meteorological data. The model training module is used to process the image data to construct an emergency scenario model. The path planning module is used to perform optimal path planning for the UAV based on the optimized meteorological data and the constructed emergency scenario model.

[0046] Thus, by optimizing the inspection path of the UAV based on the meteorological conditions and the 3D model, it can avoid inspection accidents caused by meteorology, enabling the UAV to fly in the safest and most efficient areas of the emergency scenario, always maintaining the optimal path for inspection, and eliminating the need for manual takeover, significantly improving the inspection efficiency. Brief Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for description in the embodiments or the prior art.

[0048] Figure 1 It is the schematic diagram of the UAV emergency scenario inspection system based on deep reinforcement learning of the present invention.

[0049] Figure 2 It is the schematic diagram of the environmental perception and acquisition module of the present invention.

[0050] Figure 3 It is the schematic diagram of the model training module of the present invention.

[0051] Figure 4 It is the schematic diagram of the emergency area terrain construction unit of the present invention.

[0052] Figure 5 It is the schematic diagram of the data visualization display unit of the present invention.

[0053] Figure 6 It is the schematic diagram of the path planning module of the present invention.

[0054] Figure 7 It is the step flowchart of the UAV emergency scenario inspection method based on deep reinforcement learning of the present invention.

[0055] 1 - Emergency patrol communication module, 2 - Environmental perception and acquisition module, 201 - Terrain pre - scanning unit, 202 - Light acquisition unit, 203 - Wind direction and speed acquisition unit, 204 - Weather acquisition unit, 205 - Meteorological data integration unit, 3 - Data processing module, 4 - Model training module, 401 - Emergency area terrain construction unit, 4011 - Graphic pre - processing sub - unit, 4012 - 3D reconstruction sub - unit, 4013 - Texture mapping sub - unit, 4014 - Reconstruction optimization sub - unit, 402 - Data visualization and display unit, 4021 - Time synchronization sub - unit, 4022 - Space synchronization sub - unit, 4023 - Data standardization sub - unit, 4024 - Display scaling sub - unit, 5 - Path planning module, 501 - Path preset unit, 502 - Meteorological real - time update unit, 503 - Meteorological threshold trigger unit, 504 - Patrol route switching unit, 505 - UAV current state reading unit, 506 - UAV action control unit, 507 - Emergency adjustment unit. Detailed implementation manners

[0056] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as a limitation to the present invention.

[0057] Please refer to Figures 1 to 6 , the present invention provides a UAV emergency scenario patrol system based on deep reinforcement learning, specifically including:

[0058] The emergency patrol communication module 1 is used to set up patrol communication equipment on the UAV so as to establish a real - time communication connection with the upper computer during the emergency scenario patrol;

[0059] When the UAV flies to the emergency scenario, start the patrol communication equipment set on the UAV, so as to establish a communication connection with the upper computer, so that firefighters or rescue workers can understand the disaster situation.

[0060] The environmental perception and acquisition module 2 is used to set sensors and high - precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data;

[0061] Specifically including:

[0062] The terrain pre - scanning unit 201 is used to perform terrain scanning on the emergency scenario before the emergency patrol to obtain image data for constructing an emergency scenario model;

[0063] By setting a high - precision camera and a laser scanning device on the UAV, and scanning the scenario in advance when there is no emergency, and then using the obtained image data for model construction, so that the UAV can perform path planning subsequently.

[0064] The light collection unit 202 is used to collect light intensity and light angle during emergency scene inspection to obtain light data;

[0065] During emergency scene flight inspection, the light data is obtained through the set light sensor and then used for subsequent path planning.

[0066] The wind direction and wind speed collection unit 203 is used to collect wind direction and wind speed during emergency scene inspection to obtain wind direction and wind speed data;

[0067] During emergency scene flight inspection, the wind direction and wind speed data is obtained through the set wind speed and wind direction sensors and then used for subsequent path planning.

[0068] The weather collection unit 204 is used to collect whether it is raining, whether there is lightning, and temperature and humidity conditions during emergency scene inspection to obtain weather data;

[0069] During emergency scene flight inspection, by setting a rain sensor and a networking module on the UAV, the surrounding weather can be observed, and the local real-time weather forecast can also be obtained through networking. A temperature and humidity sensor can also be set on the UAV to obtain detailed weather data, and then it is used for subsequent path planning.

[0070] The meteorological data integration unit 205 is used to integrate light data, wind direction and wind speed data, and weather data to obtain meteorological data.

[0071] Integrate all the above data for subsequent unified preprocessing.

[0072] The data processing module 3 is used to preprocess the meteorological data to obtain optimized meteorological data;

[0073] Through preprocessing, missing values can be filled, outliers can be replaced, and noise can be smoothed; the integrity and accuracy of the data can be improved, and the efficiency and accuracy of subsequent path planning can be increased.

[0074] The model training module 4 is used to process image data to build an emergency scene model;

[0075] Specifically including:

[0076] The emergency area terrain construction unit 401 is used to build the emergency scene model according to the image data;

[0077] Specifically including:

[0078] The graphic preprocessing subunit 4011 is used to preprocess the image data to obtain preprocessed image data;

[0079] By performing smoothing, contrast enhancement, brightness adjustment, and sharpening on the image, irrelevant information such as noise and speckles in the image can be removed, making the image clearer; useful information lost or weakened due to shooting conditions, transmission, etc. in the image can be restored, improving the image quality; so as to facilitate the subsequent construction of a model of the emergency scenario based on the graphic data, thereby ensuring the accuracy and efficiency of path planning.

[0080] The 3D reconstruction subunit 4012 is used to extract point clouds from the preprocessed image data and construct a three-dimensional grid model of the emergency scenario based on the point cloud data and using the Poisson reconstruction algorithm;

[0081] Extract point cloud data from the preprocessed image data; during the Poisson reconstruction process, first, the normal vector of each point in the point cloud needs to be calculated, and then it is assumed that all points are located on the surface of an unknown model. By estimating the indicator function of the model and extracting the isosurface, finally, the Marching Cubes (MC) algorithm is used to complete the surface reconstruction; after obtaining the isosurface, the MC algorithm is used to construct a triangular mesh. The MC algorithm traverses a regular voxel grid, judges each voxel to determine whether it is on the surface; connects adjacent surface voxels to form a triangular mesh, thereby completing the surface reconstruction and obtaining a three-dimensional grid model of the emergency scenario.

[0082] The texture mapping subunit 4013 is used to project the original image onto the surface of the three-dimensional grid model of the emergency scenario, and then optimize the texture seam through the Markov random field to obtain the emergency scenario model;

[0083] Image projection is the process of mapping a two-dimensional image onto the surface of a three-dimensional model. In the three-dimensional grid model of the emergency scenario, the plane projection method is used to project the original image onto the model surface according to the specified direction and angle; more realistic and rich texture information can be given to the model, thereby improving the visual fidelity of the model; using the Markov random field to optimize the texture seam can eliminate or reduce problems such as misalignment, breakage, or repetition of the texture at the seam, making the texture smoother and more continuous on the model surface.

[0084] The reconstruction optimization subunit 4014 is used to perform smoothing and hole filling on the emergency scenario model.

[0085] By performing smoothing on the model, the visual effect of the model can be improved, making it smoother and more natural; the geometric accuracy of the model can be improved, and errors caused by rough or irregular meshes can be reduced.

[0086] The data visualization and display unit 402 is used to map the meteorological data onto the emergency scenario model.

[0087] Provide an intuitive visual display of the emergency scenario model to help staff quickly understand the spatial layout and terrain features of the emergency scenario.

[0088] Specifically include:

[0089] The time synchronization subunit 4021 is used to synchronize the information of the optimized meteorological data with the time of the emergency scenario model to reduce errors.

[0090] Synchronize the acquisition time of each item of data in the optimized meteorological data with the time of the model, and update the changes according to the time synchronization, so as to facilitate the staff to accurately judge the situation of the emergency scenario.

[0091] The space synchronization subunit 4022 is used to synchronize the geographical coordinates of the optimized meteorological data to the emergency scenario model.

[0092] Mark the geographical coordinates of each item of data in the optimized meteorological data on the model, and update the changes according to the time synchronization, so as to facilitate the staff to accurately judge the situation of the emergency scenario.

[0093] The data standardization subunit 4023 is used to normalize the illumination, wind direction and speed, temperature and humidity information of the optimized meteorological data to the interval [0, 1] and display it on the emergency scenario model.

[0094] Scale the data proportionally so that it falls within the range of the interval [0, 1], and then display the normalized data items on the model, which is convenient for the staff to view uniformly and improves the judgment efficiency.

[0095] The display scaling subunit 4024 is used to add a scaling function to the displayed optimized meteorological data, and the display size can be scaled.

[0096] Adding a scaling function can change the display size of the data at any time, so that the staff can customize the settings according to their usage habits, improving the usability and flexibility.

[0097] The path planning module 5 is used to perform the optimal path planning of the unmanned aerial vehicle based on the optimized meteorological data and the constructed emergency scenario model.

[0098] Specifically include:

[0099] The path preset unit 501 is used to, when there is no emergency inspection, the staff pre-set the most efficient flight path and multiple alternative flight paths in the unmanned aerial vehicle in advance, and if there is no meteorological anomaly, fly according to the preset path.

[0100] By pre - setting the path in advance, when there is no abnormal situation, the drone will fly according to the pre - set path. This path can be explored on - site by the staff and is the optimal path under normal circumstances. Once a situation occurs, a backup path will be adopted. The inspection efficiency of the backup path is second only to that of the normal path; thus, the inspection efficiency can be greatly improved.

[0101] The meteorological real - time update unit 502 is used to update the optimized meteorological data on the emergency scenario model in real - time.

[0102] The real - time updated data can facilitate the staff to observe and judge at any time whether the drone is in a safe inspection environment.

[0103] The meteorological threshold trigger unit 503 is used to set corresponding first - level thresholds for each item of the optimized meteorological data respectively. When any data exceeds the corresponding first - level threshold, it indicates that there is a situation affecting the safety or flight efficiency of the drone. At this time, the inspection route switching unit 504 is activated to switch the inspection route.

[0104] Each item of data includes light intensity, wind speed intensity, temperature and humidity intensity, thunderstorm weather intensity, and precipitation intensity. After the data exceeds the first - level threshold, it indicates that the meteorological conditions in this area are not suitable for the safe and stable flight of the drone and will also affect the inspection efficiency; therefore, it is necessary to switch the inspection route to ensure the safe flight of the drone and the inspection with the highest efficiency.

[0105] The inspection route switching unit 504 is used to automatically switch the backup flight path of the drone after any data exceeds the first - level threshold. If the first - level threshold is still triggered after multiple flight path switches, the drone motion control unit 506 will make a micro - adjustment of the path.

[0106] When all backup paths have been switched and used, and the first - level threshold is still triggered, and there is no backup path available at this time, the drone needs to make subsequent micro - adjustments to its path so that it can inspect as far away from the dangerous area as possible under the current path condition to ensure the safety of the drone and the inspection efficiency.

[0107] The current state reading unit 505 of the drone is used to read the current flight speed, direction, and battery power of the drone for the staff to observe at any time.

[0108] Monitor the state of the drone so that the staff can understand the inspection situation of the drone and whether there are any safety problems.

[0109] The drone motion control unit 506 is used to, when the meteorological threshold is triggered again after the flight is switched, identify the specific area of the data on the emergency scenario model and control the drone to fly out of the meteorological coverage area of the data and bypass this area during subsequent inspections.

[0110] This unit can control the UAV to perform multi-directional movements in the up, down, left, right, front, and back directions, and use the emergency scenario model as a map. At this time, after the inspection by the UAV, the areas exceeding the first-level threshold have been marked on the emergency scenario model. The UAV only needs to adjust in multiple directions to avoid this area, and then avoid this area during subsequent path flights for inspection. This can ensure that the UAV always maintains the highest inspection efficiency and safe flight state under the current circumstances.

[0111] The emergency adjustment unit 507 is used to set the second-level threshold. When the first-level threshold exists in all areas of the emergency scenario model but does not exceed the second-level threshold, it controls the UAV to avoid the meteorological coverage area of the second-level threshold. When all areas exceed the second-level threshold, it controls the UAV to return.

[0112] When all areas exceed the first-level threshold, it means that there are meteorological conditions affecting the normal flight of the UAV in all areas of the emergency scenario. At this time, although it will affect flight safety and inspection efficiency, there is no risk of crashing. Therefore, in order to ensure normal communication in the emergency scenario, the UAV defaults to continue the inspection according to the current path; if all exceed the second-level threshold, it means that the meteorological data in all areas will seriously affect the flight safety and flight efficiency of the UAV, and there is a risk of crashing, and it is meaningless to continue flying. Therefore, the UAV needs to return and no longer conduct inspections; thus, through the setting of multiple paths, combined with the micro-adjustment of the UAV and threshold reminder, the highest-efficiency inspection of the emergency scenario can be carried out to the greatest extent while ensuring the flight safety of the UAV.

[0113] Please refer to Figure 7 , the present invention also provides a method for inspecting an emergency scenario of a UAV based on deep reinforcement learning, including the following steps:

[0114] S1: Set an inspection communication device on the UAV to enable real-time communication connection with the upper computer during the inspection of the emergency scenario;

[0115] S2: Set sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data;

[0116] S3: Preprocess the meteorological data to obtain optimized meteorological data;

[0117] S4: Process the image data to construct an emergency scenario model;

[0118] S5: Perform optimal path planning for the UAV based on the optimized meteorological data and the constructed emergency scenario model;

[0119] S6: The UAV starts to inspect the emergency scenario.

[0120] Among them, an inspection communication device is set on the UAV to establish a real-time communication connection with the upper computer during the inspection of emergency scenarios; sensors and high-precision cameras are set to scan and photograph emergency scenarios to obtain meteorological data and image data; the meteorological data is preprocessed to obtain optimized meteorological data; the image data is processed to construct an emergency scenario model; based on the optimized meteorological data and the constructed emergency scenario model, the optimal path planning of the UAV is carried out; the UAV is started to inspect the emergency scenario.

[0121] The above-disclosed are only one or more preferred embodiments of the present application, and the scope of the rights of the present application cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A drone emergency scenario inspection system based on deep reinforcement learning, characterized in that it includes an emergency inspection communication module, an environmental perception and acquisition module, a data processing module, a model training module, and a path planning module. The emergency inspection communication module, the environmental perception and acquisition module, the data processing module, the model training module, and the path planning module are connected in sequence; The emergency inspection communication module is used to set up inspection communication equipment on the drone to establish a real-time communication connection with the host computer during emergency scenario inspections; The environmental perception and acquisition module is used to set sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data; The data processing module is used to preprocess the meteorological data to obtain optimized meteorological data; The model training module is used to process the image data to build an emergency scenario model; The path planning module is used to perform optimal path planning for the drone based on the optimized meteorological data and the built emergency scenario model.

2. The drone emergency scenario inspection system based on deep reinforcement learning according to claim 1, characterized in that the environmental perception and acquisition module includes a terrain pre-scanning unit, a light collection unit, a wind direction and speed collection unit, a weather collection unit, and a meteorological data integration unit. The terrain pre-scanning unit, the light collection unit, the wind direction and speed collection unit, the weather collection unit, and the meteorological data integration unit are connected in sequence; The terrain pre-scanning unit is used to perform terrain scanning on the emergency scenario before emergency inspection to obtain image data for building an emergency scenario model; The light collection unit is used to collect light intensity and light angle during emergency scenario inspections to obtain light data; The wind direction and speed collection unit is used to collect wind direction and speed during emergency scenario inspections to obtain wind direction and speed data; The weather collection unit is used to collect whether it is raining, whether there is lightning, and temperature and humidity conditions during emergency scenario inspections to obtain weather data; The meteorological data integration unit is used to integrate the light data, wind direction and speed data, and weather data to obtain meteorological data.

3. The drone emergency scenario inspection system based on deep reinforcement learning according to claim 2, characterized in that the model training module includes an emergency area terrain construction unit and a data visualization display unit. The emergency area terrain construction unit and the data visualization display unit are connected; The emergency area terrain construction unit is used to build the emergency scenario model according to the image data; The data visualization display unit is used to map the meteorological data onto the emergency scenario model.

4. The drone emergency scenario inspection system based on deep reinforcement learning according to claim 3, characterized in that the emergency area terrain construction unit includes a graphic preprocessing subunit, a 3D reconstruction subunit, a texture mapping subunit, and a reconstruction optimization subunit. The graphic preprocessing subunit, the 3D reconstruction subunit, the texture mapping subunit, and the reconstruction optimization subunit are connected in sequence; The graphic preprocessing subunit is used to preprocess the image data to obtain preprocessed image data; The 3D reconstruction subunit is used to extract point clouds from the preprocessed image data and construct a three-dimensional grid model of the emergency scenario based on the point cloud data using the Poisson reconstruction algorithm; The texture mapping subunit is used to project the original image onto the surface of the three-dimensional grid model of the emergency scenario, and then optimize the texture seams through a Markov random field to obtain the emergency scenario model; The reconstruction optimization subunit is used to perform smoothing processing and hole filling on the emergency scenario model.

5. The drone emergency scenario inspection system based on deep reinforcement learning according to claim 4, wherein The data visualization and display unit includes a time synchronization subunit, a space synchronization subunit, a data normalization subunit, and a display scaling subunit, and the time synchronization subunit, the space synchronization subunit, the data normalization subunit, and the display scaling subunit are connected in sequence; The time synchronization subunit is used to synchronize the information of the optimized meteorological data with the time of the emergency scenario model to reduce errors; The space synchronization subunit is used to synchronize the geographical coordinates of the optimized meteorological data to the emergency scenario model; The data normalization subunit is used to normalize the illumination, wind direction and speed, temperature and humidity information of the optimized meteorological data to the interval [0, 1] and display it on the emergency scenario model; The display scaling subunit is used to add a scaling function to the displayed optimized meteorological data to scale the display size.

6. The drone emergency scenario inspection system based on deep reinforcement learning according to claim 5, wherein The path planning module includes a path preset unit, a meteorological real-time update unit, a meteorological threshold trigger unit, a patrol route switching unit, a current state reading unit of the drone, a drone action control unit, and an emergency adjustment unit, and the path preset unit, the meteorological real-time update unit, the meteorological threshold trigger unit, the patrol route switching unit, the current state reading unit of the drone, the drone action control unit, and the emergency adjustment unit are connected in sequence; The path preset unit is used to, when there is no emergency inspection, the staff pre-sets the most efficient flight path and multiple standby flight paths in the drone in advance, and if there is no meteorological anomaly, fly according to the preset path; The meteorological real-time update unit is used to update the optimized meteorological data on the emergency scenario model in real time; The meteorological threshold trigger unit is used to set corresponding first-level thresholds for each data of the optimized meteorological data respectively. When any data exceeds the corresponding first-level threshold, it indicates that there is a situation affecting the safety or flight efficiency of the drone. At this time, the patrol route switching unit is activated to switch the patrol route; The patrol route switching unit is used to, when any data exceeds the first-level threshold, the drone automatically switches to a standby flight path. If the first-level threshold is still triggered after multiple flight path switches, the drone action control unit performs a path micro-adjustment; The current state reading unit of the drone is used to read the current flight speed, direction and power of the drone so that the staff can observe at any time; The drone action control unit is used to identify the specific area of the data on the emergency scenario model and control the drone to fly out of the meteorological coverage area of the data and bypass the area during subsequent inspections when the meteorological threshold is triggered again after switching flights; The emergency adjustment unit is used to set a secondary threshold. When the primary threshold exists in all areas of the emergency scenario model but does not exceed the secondary threshold, it controls the drone to avoid the meteorological coverage area of the secondary threshold. When all areas exceed the secondary threshold, it controls the drone to return.

7. A method for inspecting emergency scenarios of drones based on deep reinforcement learning, using the drone emergency scenario inspection system based on deep reinforcement learning as described in claim 6, characterized in that, It includes the following steps: Set up an inspection communication device on the drone to establish a real-time communication connection with the upper computer during emergency scenario inspections; Set up sensors and high-precision cameras to scan and photograph the emergency scenario to obtain meteorological data and image data; Preprocess the meteorological data to obtain optimized meteorological data; Process the image data to build an emergency scenario model; Based on the optimized meteorological data and the built emergency scenario model, perform the optimal path planning of the drone; The drone starts to inspect the emergency scenario.

Images