Intelligent high-altitude camera cleaning method based on unmanned aerial vehicle

Abstract

The invention relates to a high-altitude camera intelligent cleaning method based on an unmanned aerial vehicle, and the method comprises the following steps: S1, setting an unmanned aerial vehicle intelligent cleaning system, S2, obtaining the omnibearing image of a to-be-cleaned high-altitude camera through a control system and a high-definition camera of the unmanned aerial vehicle, S3, introducing a multi-scale attention mechanism to improve a YOLO algorithm, carrying out the detection of smudge points of the image, and carrying out the cleaning of the smudge points of the to-be-cleaned high-altitude camera. S4, the recognized stain points are converted into three-dimensional space coordinates, the dip angle of a spray head is adjusted, and a central processing system starts a spraying system to conduct precise positioning cleaning, S5, after cleaning, the unmanned aerial vehicle is controlled to conduct all-dimensional photographing on the cleaned camera, key point detection of the stain points is conducted on the obtained picture again, and S6, if no stain point is detected, the unmanned aerial vehicle is controlled to conduct all-dimensional photographing on the camera. Controlling the unmanned aerial vehicle to return to complete cleaning; and if the smudge points are detected, all-around secondary cleaning is carried out on the smudge points until the smudge points do not exist, and cleaning is completed. The cleaning quality can be improved, the cost is saved, and the risk is reduced.

Description

Technical Field

[0001] The invention relates to an intelligent detection method based on video images, belonging to the technical field of artificial intelligence. Background Art

[0002] With the advancement of image processing technology, the price of cameras is getting lower and lower, and the application of video images is becoming more and more extensive, such as road monitoring, unmanned driving, safety detection and other industries. Although the protection of outdoor cameras is getting better and better, with the popularization of cameras, the cleaning of cameras located at high altitudes outdoors has become an increasingly urgent problem. Due to the high installation position, high-altitude cameras have problems such as high difficulty in cleaning, low cleaning efficiency, difficult to ensure cleaning quality, high maintenance costs, and high safety risks for cleaning personnel.

[0003] The rapid development of low-altitude economy, space technology and artificial intelligence technology has provided technical and equipment support for the high-quality and rapid cleaning of high-altitude cameras. Cleaning systems, equipment, patents, etc. based on drones are constantly being developed, such as "a multi-rotor drone cleaning system", "a method and system for automatically cleaning photovoltaic electric fields based on drones", "intelligent drone vertical cleaning system", "image transmission and processing methods for drones used for high-altitude curtain wall cleaning", etc. These cleaning systems are suitable for high-altitude curtain walls, photovoltaic power plants, etc.; devices suitable for high-altitude cameras, such as "a cleaning system for cleaning high-altitude cameras", have also been developed. High-altitude cameras are different from large targets such as curtain walls and photovoltaic power plants. At present, there are still problems such as small and non-planar cleaning targets, difficult stain point detection, and difficult to ensure cleaning quality.

[0004] The present invention aims to provide an efficient, accurate and intelligent cleaning method and system suitable for high-altitude cameras based on drones and video images, combined with deep learning methods and three-dimensional spatial positioning technology. Summary of the invention

[0005] In view of the current problems of high-altitude camera cleaning, such as small and non-planar cleaning targets, difficulty in detecting stains, and difficulty in ensuring cleaning quality, the present invention provides an efficient, accurate, and intelligent high-altitude camera cleaning method and system based on drones. The method uses the control system and high-definition camera of the drone, combined with the improved YOLO algorithm, to detect small non-planar targets in the cleaning of high-altitude cameras, namely stains; uses the control system and three-dimensional spatial positioning algorithm of the drone to achieve accurate and focused cleaning of stains, and monitors and evaluates the cleaning effect. The present invention can improve cleaning quality, save costs, and reduce risks.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows: a high-altitude camera intelligent cleaning method based on a drone, the method comprising the following steps: S1: Setting up a drone intelligent cleaning system, wherein the drone intelligent cleaning system includes a drone, a high-definition camera, a spray system, and a central processing system, wherein the high-definition camera, the spray system, and the central processing system are all mounted on the drone. S2: Using the control system and high-definition camera of the drone, obtain all-round pictures of the high-altitude camera to be cleaned; S3: Introduce a multi-scale attention mechanism to improve the YOLO algorithm, and detect stain points in the image based on the improved YOLO algorithm; S4: The identified stain points are converted into three-dimensional spatial coordinates, the nozzle inclination is adjusted, and the central processing system starts the spray system for precise positioning and cleaning; S5: After cleaning, the drone is controlled to take all-round photos of the cleaned camera, and the key points of the stain points are detected again on the acquired images; S6: If no stain is detected, the drone is controlled to return and the cleaning is completed; if a stain is detected, the stain is cleaned again in all directions, and after cleaning, photos are taken and stain detection is performed until there is no stain, and the cleaning is completed.

[0007] Among them, S1: In the UAV intelligent cleaning system, Drone: Used to build an intelligent sprinkler system. High-definition cameras, drone control systems, central processing systems, sprinkler systems, etc. are all carried on drones.

[0008] High-definition camera: installed on the top of the drone to capture the high-altitude camera images that need to be cleaned; Spray system: used to clean the high-altitude camera, including cleaning fluid, a container for holding the cleaning fluid, a spray head, a device for monitoring the capacity of the cleaning fluid, etc.; Central processing system: used to control the entire drone cleaning system, including intelligent recognition module and control module. The intelligent recognition module is used to process images, detect the stains on the camera that need to be cleaned, and determine the parts that need to be cleaned; the control module includes drone control system module, camera control system module, spray system control module and information transmission module, etc., which are used to control the flight status of the drone, record status data, camera shooting, nozzle angle adjustment and spray system status, and transmit drone flight demand information to the drone control system.

[0009] S2: Use the drone’s control system and high-definition camera to obtain a full range of images of the high-altitude camera to be cleaned, including: S2-1 Use the control system of the drone to obtain the three-dimensional spatial coordinates of the high-altitude camera to be cleaned. S2-2 uses the control system and high-definition camera of the drone to control the drone to take pictures around the high-altitude camera that needs to be cleaned. When taking pictures, the previous picture and the next picture must overlap by more than 60%, and a full range of images of the high-altitude camera that needs to be cleaned can be obtained; S2-3 uses the positioning system in the control system of the drone to determine the flight parameters of the drone and the position of the camera when each picture is taken.

[0010] S3: Introduce a multi-scale attention mechanism to improve the YOLO algorithm, and detect stain points in the image based on the improved YOLO algorithm, including: S3-1 first performs image enhancement processing on the captured image; S3-2 introduces a multi-scale attention mechanism to improve the YOLO algorithm and improve the YOLO algorithm; S3-21 The overall framework of the YOLO network model includes four parts: Input, Backbone, Neck, and Head. S3-22 The multi-scale attention mechanism EMA is introduced in the YOLO network. EMA groups the channel dimensions and uses three parallel paths to extract the attention weights of the grouped feature maps to improve the attention to small target features. S3-3 uses the improved YOLO algorithm to detect the key points of stains on the image and output the image coordinates of the camera and each stain point on the camera; S4: Convert the identified stain points into three-dimensional space coordinates, adjust the nozzle inclination, and the central processing system starts the spray system for precise positioning and cleaning, which also includes: S4-1 locates the stain points identified by YOLO; S4-2 controls the drone to fly to the location when the picture was taken according to the flight parameters of the drone recorded when the picture was taken; S4-3 Adjust the inclination angle of the nozzle according to the three-dimensional coordinates of the stain points identified by YOLO; S4-4: for other identified stain points, repeat the steps from S4-1 to S4-4 to clean the detected stain points one by one; S4-5 performs normal cleaning on the entire camera.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides an efficient and intelligent cleaning method for a high-altitude camera. By introducing a multi-scale attention mechanism EMA to improve the YOLO algorithm, the recognition ability of small targets is enhanced, and the problem of difficulty in detecting non-planar small targets (stain spots) in the cleaning of high-altitude cameras is solved; through the unmanned aerial vehicle positioning system and the three-dimensional space positioning algorithm, the appropriate nozzle angle is determined, the nozzle is adjusted, and the focus cleaning of the stain spots is achieved, and the precise detection and precise cleaning quality of the parts that need to be focused on cleaning are achieved, thereby improving the cleaning quality; the present invention can also evaluate the cleaning effect, achieve precise control of the cleaning quality, and ensure the cleaning effect. The present invention can improve the cleaning quality of high-altitude cameras, save costs, and reduce risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart of the present invention, Figure 2 This is a diagram of drone photography and cleaning. Figure 3 This is the YOLO network architecture diagram. Figure 4 It is an EMA diagram. DETAILED DESCRIPTION

[0013] In order to deepen the understanding of the present invention, the specific implementation of the present invention is further explained below with reference to the accompanying drawings.

[0014] Embodiment 1: like Figure 1 As shown, the present invention provides a high-altitude camera intelligent cleaning method based on a drone, comprising the following steps: S1: Setting up a drone intelligent cleaning system, wherein the drone intelligent cleaning system includes a drone, a high-definition camera, a spray system, and a central processing system, wherein the high-definition camera, the spray system, and the central processing system are all mounted on the drone. Drone: Used to build an intelligent sprinkler system. High-definition cameras, drone control systems, central processing systems, sprinkler systems, etc. are all carried on drones.

[0015] High-definition camera: installed on the top of the drone to capture the high-altitude camera images that need to be cleaned; Spray system: used to clean the high-altitude camera, including cleaning fluid, a container for holding the cleaning fluid, a spray head, a device for monitoring the capacity of the cleaning fluid, etc.; Central processing system: used to control the entire drone cleaning system, including intelligent recognition module and control module. The intelligent recognition module is used to process images, detect the stains on the camera that need to be cleaned, and determine the parts that need to be cleaned; the control module includes drone control system module, camera control system module, spray system control module and information transmission module, etc., which are used to control the flight status of the drone, record status data, camera shooting, nozzle angle adjustment and spray system status, and transmit drone flight demand information to the drone control system.

[0016] S2: Use the drone’s control system and high-definition camera to obtain a full range of images of the high-altitude camera to be cleaned, including: S2-1 Use the drone control system to obtain the three-dimensional spatial coordinates of the high-altitude camera to be cleaned. First, use the drone control system to control the drone to hover directly below the camera to be cleaned, and use the positioning function of the drone control system to obtain the coordinates of the center point of the camera to be cleaned. ; Then control the drone to hover in a parallel position to the center of the camera that needs to be cleaned, and obtain the coordinates of the center point of the camera that needs to be cleaned according to the positioning system of the drone ; Locate the 3D coordinates of the center point of the camera that needs to be cleaned ; S2-2 uses the control system and high-definition camera of the drone to control the drone to take pictures around the high-altitude camera that needs to be cleaned. When taking pictures, the previous picture and the next picture must overlap by more than 60%, and a full range of images of the high-altitude camera that needs to be cleaned can be obtained; S2-3 uses the positioning system in the drone control system to obtain the flight parameters of the drone when each picture is taken and the position of the high-definition camera on the drone when shooting The flight parameters include the flight altitude, the three-dimensional coordinates of the drone , UAV attitude parameters, record the focal length of the camera , the width and height of the camera frame , , shooting angle , and correspond one to one with the captured images. See the shooting diagram for Figure 2 ; S3: Introduce a multi-scale attention mechanism to improve the YOLO algorithm, and detect stain points in the image based on the improved YOLO algorithm, including: S3-1 first performs image enhancement processing on the captured image. The specific formula for image enhancement is as follows: (1), In the formula, is a point in the image Gray value of express The corresponding enhanced pixel value, Is the image The local average value centered at the local template size is , see formula (2) for specific calculation, is the local standard deviation, and the specific calculation is shown in formula (3). (2) , In the formula, is the number of pixels, is the local template size, is the midpoint of the image The pixel value at Value from arrive , Value from arrive , (3).

[0017] S3-2 improves the YOLO algorithm, including: S3-2-1 The overall framework of the YOLO network model includes four parts: Input, Backbone, Neck and Head. For details, see Figure 3Input receives channel images of different resolutions; Backbone is the backbone of the network, which is responsible for extracting features of images at multiple scales using convolutional neural networks and outputting feature vectors of different dimensions. Conv is the convolution layer. The C3k2 module is a faster CSP (Cross Stage Partial Networks) module variant introduced by YOLO11 in the Backbone part. It uses two parallel convolution layers (two C3k) to improve the efficiency of data processing and increase the depth of feature extraction. The C2PSA module is a new feature extraction and processing module with an attention mechanism. It introduces the pyramid slice attention (PSA) mechanism to optimize the effectiveness of the model in multi-scale feature processing. The SPPF module performs maximum pooling at different scales and connects the results to capture features at multiple spatial scales. The Neck part is responsible for the preliminary feature extraction and feature fusion of the feature vectors received from Backbone, which involves upsampling of feature maps at different levels (used to sample smaller feature maps to the same size as other feature maps) and concatenation (mainly used to splice and combine feature maps of different levels and scales to improve the accuracy and efficiency of target detection). The Head part is responsible for the final target detection and classification tasks, identifying targets of different sizes based on different detection layer dimensions, that is, generating bounding boxes and classifying objects in these boxes. The Detect class defines the detection head that outputs bounding box coordinates and class probabilities. S3-2-2 Introducing the multi-scale attention mechanism in the YOLO network, see Figure 4 , Figure 4 This is a schematic diagram of EMA. The EMA module combines channel and spatial information to preserve information in the channel dimension and reduce the computational burden. It also groups the channel dimension into multiple sub-features so that the spatial semantic features are evenly distributed in each feature group. First, the channel dimension is grouped , using three parallel paths to extract the attention weights of the grouped feature maps, including two 1x1 convolutions and a single 3x3 convolution; then encoding the global spatial information through two-dimensional global average pooling and Softmax function, and using matrix dot product operations to aggregate cross-spatial information in different spatial dimensions; finally, after Sigmod activation function and adaptive reweighting, the model is provided with global context information, enabling EMA to generate better pixel-level attention on deep feature maps and improve attention to small target features.

[0018] S3-3 uses the improved YOLO algorithm to detect the key points of stains on the image and output the image coordinates of the camera and each stain point on the camera. , They are respectively in the upper left corner of the image , Coordinates of its width and height .

[0019] S4: The identified stain points are converted into three-dimensional spatial coordinates, the nozzle inclination is adjusted, and the central processing system starts the spray system for precise positioning and cleaning, including: S4-1 locates the stain points on the image identified by YOLO and converts them into three-dimensional space coordinates, as follows: S4-1-1 Image coordinates of the camera to be cleaned according to YOLO recognition , the image coordinates of the stain points to be cleaned , calculate the offset of the stain point relative to the center point of the camera to be cleaned on the image and , (4) (5) In the formula, is the horizontal coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the vertical coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, S4-1-2 Based on the image of the identified stain, determine the distance between the high-definition camera on the drone and the high-altitude camera to be cleaned when taking the image , (6) In the formula, is the distance between the high-definition camera on the drone and the camera to be cleaned, R is the radius of the earth, and its value is 6371.393km; is the longitude coordinate of the HD camera, is the latitude coordinate of the HD camera, is the longitude coordinate of the center point of the camera to be cleaned, is the latitude coordinate of the center point of the camera to be cleaned, S4-1-3 Calculate the width and height of the captured image , , (7), (8), In the formula, is the width of the captured image, is the height at which the image was taken, is the focal length of the camera, , are the width and height of the picture, S4-1-4 Calculate the actual offset distance of the stain point relative to the center point of the camera and , (9) (10) In the formula, is the horizontal coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the vertical coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the actual horizontal coordinate offset distance of the stain point relative to the center point of the camera to be cleaned. is the actual vertical offset distance of the stain point relative to the center point of the camera to be cleaned. S4-1-5 Calculate the three-dimensional coordinates of the camera , calculate the spatial three-dimensional coordinates of the stain point , (11) (12) (13) in, is the longitude coordinate of the center point of the camera to be cleaned, is the latitude coordinate of the center point of the camera to be cleaned, is the elevation coordinate of the center point of the camera to be cleaned, is the longitude coordinate of the smudge point, is the latitude coordinate of the smudge point, is the elevation coordinate of the stain point.

[0020] S4-3 Adjust the angle between the nozzle and the ground plane according to the three-dimensional coordinates of the stain points identified by YOLO , as follows: S4-3-1 Determine the coordinates of the camera on the drone when the picture was taken based on the marked picture information , calculate the coordinates of the nozzle : is the longitude coordinate of the sprinkler head, is the latitude coordinate of the sprinkler, is the elevation coordinate of the nozzle, is the longitude coordinate of the HD camera, is the latitude coordinate of the HD camera, is the elevation coordinate of the HD camera, is the horizontal distance between the HD camera and the nozzle, is the height difference between the two, which is a fixed value; the nozzle is on the right side of the camera (east), It is positive when it is above the camera (north), negative when it is below the camera, ARC = 6371.393*1000.

[0021] S4-3-2 Calculate the horizontal distance between the nozzle and the stain point , (17) In the formula, is the horizontal distance between the nozzle and the stain point, R is the radius of the earth, and its value is 6371.393km; S4-3-3 According to the positioning information of the nozzle in the spray system in the UAV control system and the positioning information of the stain point, the angle of the nozzle is adjusted so that it can be aimed at the stain point for fixed-point and focused cleaning; (18) In the formula, is the angle between the nozzle and the ground plane, It is the vertical height difference between the camera to be cleaned and the nozzle. It is the stain spot to be cleaned. It is the horizontal distance between the nozzle and the stain point.

[0022] S4-4: for other identified stain points, repeat the steps from S4-1 to S4-4 to clean the detected stain points one by one; S4-5 performs normal cleaning on the entire camera.

[0023] S5: After cleaning, the drone is controlled to take all-round photos of the cleaned camera, and the key points of the stain points are detected again on the acquired images; S6: If no stain is detected, the drone is controlled to return and the cleaning is completed; if a stain is detected, the stain is cleaned again in all directions, and after cleaning, photos are taken and stain detection is performed until there is no stain, and the cleaning is completed.

[0024] Embodiment 2: Provided is a high-altitude camera intelligent cleaning system based on a drone, comprising: The drone is configured as: a drone equipped with an intelligent spraying system, wherein a high-definition camera, a drone control system, a central processing system, a spraying system, etc. are all carried on the drone; A high-definition camera, which is configured as follows: the high-definition camera is used to capture the high-altitude camera image that needs to be cleaned; The drone control system is configured to: control the drone status, such as flying, hovering, etc., and record flight status data, such as accurate positioning information, flight attitude, etc.; A spray system configured to: be used to clean the high-altitude camera, including a cleaning liquid, a container for holding the cleaning liquid, a spray head, a device for monitoring the capacity of the cleaning liquid, etc.; The central processing system is configured to: control the entire drone cleaning system, including an intelligent recognition module and a control module. The intelligent recognition module is used to process images, detect stains on the camera that need to be cleaned, and determine the areas that need to be cleaned; the control module is used to control the camera shooting, control the start of the spray system, control the adjustment of the spray head angle, and control the drone flight demand information transmitted to the drone control system; The proposed system can be implemented in other ways. For example, the system embodiment described above is only illustrative, and the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0025] Example 3: This embodiment also provides an intelligent cleaning device, including: a drone, a high-definition camera, a cleaning box, a nozzle, a central processing unit, a memory or multiple computer programs; wherein the processor is connected to the drone, the high-definition camera, the cleaning box, and the nozzle, and the above one or more computer programs are stored in the central processing unit. When the device is running, the central processing unit executes one or more computer programs to enable the device to execute the method described in the above embodiment one.

[0026] The method in Example 1 can be performed by combining hardware and software modules in a processor. The software module can be located in a storage medium mature in the art such as a central processing unit, a flash memory, a programmable read-only memory, etc. The software module, combined with its hardware, completes the steps of the above method. To avoid repetition, it will not be described in detail here.

[0027] It should be noted that the above embodiments are not intended to limit the protection scope of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. A high-altitude camera intelligent cleaning method based on an unmanned aerial vehicle, characterized in that: The method comprises the following steps, S1: Setting up a drone intelligent cleaning system, wherein the drone intelligent cleaning system includes a drone, a high-definition camera, a spray system, and a central processing system, wherein the high-definition camera, the spray system, and the central processing system are all mounted on the drone. S2: Use the drone's control system and high-definition camera to obtain a full range of images of the high-altitude camera to be cleaned. S3: Introduce a multi-scale attention mechanism to improve the YOLO algorithm, and detect stain points in the image based on the improved YOLO algorithm. S4: The identified stain points are converted into three-dimensional spatial coordinates, the nozzle inclination is adjusted, and the central processing system starts the spray system for precise positioning and cleaning. S5: After cleaning, the drone is controlled to take all-round photos of the cleaned camera, and the key points of the stains are detected again on the acquired images. S6: If no stain is detected, the drone is controlled to return and the cleaning is completed; If stains are detected, the stains will be cleaned again in all directions. After cleaning, photos will be taken and stains will be detected until there are no stains and the cleaning is completed.

2. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 1, characterized in that: In the UAV intelligent cleaning system, Drones: used to build intelligent sprinkler systems. High-definition camera: installed on the top of the drone to capture the high-altitude camera images that need to be cleaned; Spray system: used to clean the high-altitude camera, including cleaning fluid, a container for the cleaning fluid, a spray head, and a device for monitoring the capacity of the cleaning fluid. Central processing system: used to control the entire drone cleaning system, including intelligent recognition module and control module. The intelligent recognition module is used to process images, detect stains on the camera that needs to be cleaned, and determine the parts that need to be cleaned; the control module includes drone control system module, camera control system module, spray system control module and information transmission module, which are used to control the flight status of the drone, record status data, camera shooting, nozzle angle adjustment and spray system status, and transmit drone flight demand information to the drone control system.

3. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 1, characterized in that: In step S2, the control system and high-definition camera of the drone are used to obtain a full range of images of the high-altitude camera to be cleaned, including: S2-1 Use the drone control system to obtain the three-dimensional spatial coordinates of the high-altitude camera to be cleaned. First, use the drone control system to control the drone to hover directly below the camera to be cleaned, and use the positioning function of the drone control system to obtain the coordinates of the center point of the camera to be cleaned. ; Then control the drone to hover in a parallel position to the center of the camera that needs to be cleaned, and obtain the coordinates of the center point of the camera that needs to be cleaned according to the positioning system of the drone ; Locate the 3D coordinates of the center point of the camera that needs to be cleaned ; S2-2 uses the control system and high-definition camera of the drone to control the drone to take pictures around the high-altitude camera that needs to be cleaned. When taking pictures, the previous picture and the next picture must overlap by more than 60%, and a full range of images of the high-altitude camera that needs to be cleaned can be obtained; S2-3 uses the positioning system in the drone control system to obtain the flight parameters of the drone when each picture is taken and the position of the high-definition camera on the drone when shooting The flight parameters include the flight altitude, the three-dimensional coordinates of the drone , UAV attitude parameters, record the focal length of the camera , the width and height of the camera frame , , shooting angle , and correspond one to one with the captured images.

4. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 1, characterized in that: In step S3, a multi-scale attention mechanism is introduced to improve the YOLO algorithm, and stain points are detected on the image based on the improved YOLO algorithm, including: S3-1 first performs image enhancement processing on the captured images. S3-2 introduces a multi-scale attention mechanism to improve the YOLO algorithm. S3-3 uses the improved YOLO algorithm to detect the key points of stains in the image and output the image coordinates of the camera and each stain point on the camera.

5. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 1, characterized in that: S4: The identified stain points are converted into three-dimensional spatial coordinates, the nozzle inclination is adjusted, and the central processing system starts the spray system for precise positioning and cleaning, including: S4-1 locates the stain points on the image identified by YOLO and converts them into three-dimensional space coordinates; S4-2 controls the drone to fly to the location when the picture was taken according to the flight parameters of the drone recorded when the picture was taken; S4-3 Adjust the inclination angle of the nozzle according to the three-dimensional coordinates of the stain points identified by YOLO; S4-4: for other identified stain points, repeat the steps from S4-1 to S4-4 to clean the detected stain points one by one; S4-5 performs normal cleaning on the entire camera.

6. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 4 is characterized in that: S3-1 first performs image enhancement processing on the captured image. The specific details of image enhancement are as follows: （1） In the formula, is a point in the image Gray value of express The corresponding enhanced pixel value, Is the image The local average value centered at the local template size is , see formula (2) for specific calculation, is the local standard deviation, see Notice (3) for the specific calculation, (2) ， In the formula, is the number of pixels, is the local template size, is the midpoint of the image The pixel value at Value from arrive , Value from arrive , （3）。 7. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 4 is characterized in that: S3-2 improves the YOLO algorithm, including: S3-2-1 The overall framework of the YOLO network model includes four parts: input, backbone network, neck and head. The input receives channel images of different resolutions. The backbone network uses a convolutional neural network to extract features from the image and output feature vectors of different dimensions. The neck is responsible for performing preliminary feature extraction and feature fusion to enhance feature extraction of the feature vectors received from the backbone network. The head identifies targets of different sizes based on different detection layer dimensions. S3-2-2 YOLO network introduces a multi-scale attention mechanism EMA, which groups the channel dimensions and uses three parallel paths to extract the attention weights of the grouped feature maps, including two Convolution and single Convolution; then use two-dimensional global average pooling and Softmax function to encode global spatial information, and use matrix dot product operation to aggregate cross-spatial information in different spatial dimensions; finally, after Sigmod activation function and adaptive reweighting, the model is provided with global context information, so that EMA can generate better pixel-level attention on deep feature maps and improve the focus on small target features. S3-3 uses the improved YOLO algorithm to detect the key points of the stain points on the image and output the image coordinates of the camera to be cleaned And the image coordinates of the stain points to be cleaned ,in, is the normalized horizontal coordinate of the camera to be cleaned relative to the upper left corner of the image, is the normalized ordinate of the camera to be cleaned relative to the upper left corner of the image, is the width of the camera to be cleaned, is the height of the camera to be cleaned, is the normalized horizontal coordinate of the stain point relative to the upper left corner of the image, is the normalized ordinate of the stain point relative to the upper left corner of the image, is the width of the stain, is the height of the stain point.

8. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 5 is characterized in that: S4-1 locates the stain points on the image identified by YOLO and converts them into three-dimensional space coordinates, as follows: S4-1-1 Image coordinates of the camera to be cleaned according to YOLO recognition , the image coordinates of the stain points to be cleaned , calculate the offset of the stain point relative to the center point of the camera to be cleaned on the image and , (4)， (5)， In the formula, is the horizontal coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the vertical coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, S4-1-2 Based on the image of the identified stain, determine the distance between the high-definition camera on the drone and the high-altitude camera to be cleaned when taking the image , (6)， In the formula, is the distance between the high-definition camera on the drone and the camera to be cleaned, R is the radius of the earth, and its value is 6371.393km; is the longitude coordinate of the HD camera, is the latitude coordinate of the HD camera, is the longitude coordinate of the center point of the camera to be cleaned, is the latitude coordinate of the center point of the camera to be cleaned, S4-1-3 Calculate the width and height of the captured image , , （7）， （8）， In the formula, is the width of the captured image, is the height at which the image was taken, is the focal length of the camera, , are the width and height of the picture, S4-1-4 Calculate the actual offset distance of the stain point relative to the center point of the camera and , (9)， (10)， In the formula, is the horizontal coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the vertical coordinate offset of the stain point relative to the center point of the camera to be cleaned on the image, is the actual horizontal coordinate offset distance of the stain point relative to the center point of the camera to be cleaned. is the actual vertical offset distance of the stain point relative to the center point of the camera to be cleaned. S4-1-5 Calculate the three-dimensional coordinates of the camera , calculate the spatial three-dimensional coordinates of the stain point , (11)， (12)， (13)， in, is the longitude coordinate of the center point of the camera to be cleaned, is the latitude coordinate of the center point of the camera to be cleaned, is the elevation coordinate of the center point of the camera to be cleaned, is the longitude coordinate of the smudge point, is the latitude coordinate of the smudge point, is the elevation coordinate of the stain point.

9. The intelligent cleaning method for a high-altitude camera based on an unmanned aerial vehicle according to claim 5, characterized in that: S4-3 Adjust the angle between the nozzle and the ground plane according to the three-dimensional coordinates of the stain points identified by YOLO , as follows: S4-3-1 Determine the coordinates of the camera on the drone when the picture was taken based on the marked picture information , calculate the coordinates of the nozzle : is the longitude coordinate of the sprinkler head, is the latitude coordinate of the sprinkler, is the elevation coordinate of the nozzle, is the longitude coordinate of the HD camera, is the latitude coordinate of the HD camera, is the elevation coordinate of the high-definition camera, is the horizontal distance between the HD camera and the nozzle, is the height difference between the two, which is a fixed value; the nozzle is on the right side of the camera (east), It is positive when it is above the camera (north), and negative when it is below the camera. ARC = 6371.393*1000. S4-3-2 Calculate the horizontal distance between the nozzle and the stain point , (17)， In the formula, is the horizontal distance between the nozzle and the stain point, R is the radius of the earth, and its value is 6371.393km; S4-3-3 According to the positioning information of the nozzle in the spray system in the UAV control system and the positioning information of the stain point, the angle of the nozzle is adjusted so that it can be aimed at the stain point for fixed-point and focused cleaning; (18) In the formula, is the angle between the nozzle and the ground plane, It is the vertical height difference between the camera to be cleaned and the nozzle. It is the stain spot to be cleaned. It is the horizontal distance between the nozzle and the stain point.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instruction is executed by the processor, an intelligent cleaning method for a high-altitude camera based on a drone as described in any one of claims 1-9 is implemented.

Images