Aircraft route task control and execution system

Abstract

The invention provides an aircraft route task control and execution system, and the system comprises the steps: 1, constructing a three-dimensional map model, placing a mesh model above a 3D Gaussian sputtering model on a three-dimensional map engine, carrying out the nesting of the mesh model and the 3D Gaussian sputtering model, obtaining a mixed model of the mesh model and the 3D Gaussian sputtering model, bringing a shot picture into the mixed model, and carrying out the training, obtaining a three-dimensional map with three-dimensional coordinate data required by flight of the flight route; step 2, selecting a plurality of points on the three-dimensional map to be set as waypoints, and connecting the plurality of waypoints to form a route; step 3, endowing a route function to the waypoint; 4, simulating flight according to the waypoints selected in the step 2 and the step 3 and the functions of the waypoints; and 5, the aircraft receives a flight instruction, flies to the waypoint meeting the flight effect confirmed in the step 4, and executes a corresponding waypoint function at the same time.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicles, and in particular relates to an aircraft route mission control and execution system. Background Art

[0002] The drone mission management platform is a software solution based on cloud computing technology that provides a range of services from mission planning to flight control data processing and analysis. Such a platform aims to simplify drone operation processes, improve work efficiency, and ensure flight safety and compliance.

[0003] However, the drone mission management platform in the existing technology has the following problems: 1. The product is not easy to use, the operation steps are complicated, and it is not easy for ordinary users to operate it; 2. The product is based on a two-dimensional plane map, and only the longitude and latitude coordinates are provided when setting the waypoints. The altitude coordinates rely on the switching of the virtual perspective, and it is impossible to set high-precision waypoints; 3. The product operation results are not optimistic, and the flight effect of the waypoints cannot be verified.

[0004] At the same time, the UAV mission management platform in the existing technology is a general product, which can only fly to waypoints and perform waypoint functions at waypoints. In other words, it is just for flying and cannot solve practical application problems, such as detecting safety lines of buildings and directing ground traffic. Therefore, its functions are relatively single and cannot meet the diverse application needs in actual scenarios. Summary of the invention

[0005] The present invention provides an aircraft route mission control and execution system. By constructing a hybrid model on a map engine, a three-dimensional map with coordinate information is obtained, and waypoint setting is easier. At the same time, the waypoint flight effect can be verified through simulated flight, and the operation result is controllable.

[0006] Other purposes and advantages of the present invention can be further understood from the technical features disclosed in the present invention.

[0007] In order to achieve one or part or all of the above purposes or other purposes, a technical solution of the present invention provides: An aircraft route mission control and execution system, characterized in that it includes the following steps: Step 1: construct a three-dimensional map model, and on a three-dimensional map engine, place a mesh model on top of a 3D Gaussiansplatting model for nesting to obtain a hybrid model of a mesh model and a 3D Gaussian splatting model, and drive the placement of the mesh model and the 3D Gaussian splatting model together through the three-dimensional map engine, and bring the captured image into the hybrid model for training to obtain a three-dimensional map with three-dimensional coordinate data required for route flight; Step 2: select a number of points on the three-dimensional map to set as waypoints, and record the corresponding coordinate position information of the waypoints, and connect a number of the waypoints to form a route; Step 3: assign functions to the waypoints; Step 4: perform simulated flight according to the waypoints and waypoint functions selected in Steps 2 and 3, simulate the flight of the aircraft to each waypoint, and execute the flight effect of each waypoint, and compare the simulated flight effect with the actual required flight effect, and verify and modify the waypoints and waypoint functions in Step 3; Step 5: The aircraft receives the flight instruction, flies to the waypoints that are confirmed to meet the flight effect in Step 4, and executes the corresponding waypoint functions at the same time. The beneficial effect of this technical solution is that by constructing a hybrid model, not only can the 3D map be beautified to make it more realistic and fit the real scene, but also the coordinates can be marked for the 3D map, so that users can directly select on the 3D map when setting waypoints, simplifying the difficulty of waypoint selection and making the operation easier. At the same time, simulated flight is used to verify the waypoints and waypoint functions, making the operation results controllable and reducing the risk of accidents during flight.

[0008] In the step 1, the mesh model and the 3D Gaussian splatting model are nested in equal proportions.

[0009] The image with coordinate information generated by the mesh model in the hybrid model in step 1 is made transparent, and only the three-dimensional grid graphics with coordinate information are retained. The three-dimensional grid graphics are overlaid on the image generated by the 3D Gaussiansplatting model training to obtain a three-dimensional map with three-dimensional coordinate data.

[0010] Waypoint functions include: taking photos, changing the aircraft's yaw angle and setting the yaw angle value, changing the gimbal pitch angle and setting the gimbal pitch angle value, camera zoom and setting the zoom value, camera focus, switching lenses, taking photos, starting recording, and ending recording.

[0011] The way to assign functions to waypoints is to select points for real-scene functions, which mainly includes the following steps: S3.1: real-time video streaming, showing the picture taken by the current camera of the aircraft; S3.2: for each waypoint selected in step 2 in turn, based on the picture taken by the current camera and according to the user's visual requirements, determine whether the shooting position and angle of the aircraft at the current waypoint meet the user's visual requirements. When the shooting position and angle of the aircraft do not meet the requirements, adjust the shooting position and angle of the aircraft, and adjust the three-dimensional coordinates of the current waypoint; S3.3: assign corresponding functions to each adjusted waypoint.

[0012] In step S3.2, the method for adjusting the three-dimensional coordinates of the current waypoint is to send an adjustment command to the aircraft through the controller. The aircraft adjusts the corresponding flight posture to obtain a shooting position and shooting angle that meet the requirements. The position after the aircraft posture is adjusted is the corrected three-dimensional coordinates of the waypoint.

[0013] The control instructions received by the aircraft are sent to the airport based on the instructions issued by the keyboard or recognized by the voice recognition system. After the airport receives the instructions, it controls the aircraft to execute the corresponding instructions.

[0014] When simulating the flight effect, step 4 calculates the height and angle of the aircraft's flight based on the position information of each point of the aircraft on the three-dimensional map, and realizes the flight following effect from a first-person perspective.

[0015] In step 3, the waypoints are assigned functions in a simulated function point selection manner, and corresponding functions are assigned to each waypoint in turn based on the waypoints selected in step 2.

[0016] The aircraft route mission control and execution system also includes: under the real-scene function point selection mode, the aircraft flies to the adjusted waypoint position multiple times and takes photos; the photos taken multiple times are sent to the artificial intelligence visual algorithm for training to build an application model; after receiving the flight command, the aircraft flies to the waypoint to take photos, and uses the constructed application model to compare and analyze the photos taken at the waypoint, find out the abnormal points in the picture and output a decision plan.

[0017] Building the application model also includes: data set preparation, using a generative adversarial network to perform data enhancement on the pictures taken by the aircraft, the generator of the generative adversarial network randomly samples a vector from a latent space as the input of the generator, the generator network receives the latent vector as input, and gradually generates an image data sample through the U-NET convolutional neural network, and the generated sample and the real sample are sent to the discriminator network for discrimination; the generator continuously generates data samples and sends them to the discriminator for identification until the generator generates sufficiently realistic samples so that the discriminator cannot accurately distinguish the generated samples from the real samples, at which point the generated samples are put into the data set and sent to the artificial intelligence vision algorithm for training.

[0018] The lightweight convolutional coding network, the backbone network of the artificial intelligence vision algorithm, is improved. A hierarchical attention network is embedded in the lightweight convolutional coding network to extract image features in depth. The hierarchical attention network adopts a dynamic convolutional coding module for feature extraction, and uses small convolution kernels in shallow networks and large convolution kernels in deep networks.

[0019] The loss function used by the artificial intelligence vision algorithm for balance training is: ; Where N represents the number of detection layers, represents the classification loss, represents the bounding box regression loss, is the confidence loss, , , Represents the weights assigned to the three losses.

[0020] Compared with the prior art, the beneficial effects of the present invention mainly include: 1. Through the nested hybrid model, the three-dimensional map can be beautified to make the three-dimensional map effect more realistic and fit the real scene, and the coordinates can be marked for the three-dimensional map, so that the user can directly select on the three-dimensional map when setting the waypoint, simplifying the difficulty of waypoint selection and making the operation easier. At the same time, simulated flight is used to verify the waypoints and waypoint functions, so that the operation results are controllable and the accidental risks during the flight are reduced.

[0021] 2. In the process of selecting points in real scenes, multiple photos can be taken at the waypoint location, and the photos taken can be trained with artificial intelligence visual algorithms to obtain an application model for the waypoint, mainly a defect application model for the waypoint. Therefore, in daily patrols, when the aircraft flies to the waypoint to take photos and inputs the photos into the application model, the waypoint can be actually applied to solve corresponding problems, such as discovering defects at the waypoint and giving decision-making plans, enriching the functions of the aircraft route mission control and execution system.

[0022] In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the specific embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 The figure is a flow chart of the system of the present invention executing a flight mission.

[0025] Figure 2 The figure is a flow chart of a hybrid model technology constructed in the present invention. DETAILED DESCRIPTION

[0026] The above and other technical contents, features and effects of the present invention will be clearly presented in the following detailed description of a preferred embodiment with reference to the drawings. The directional terms mentioned in the following embodiments, such as up, down, left, right, front or back, etc., are only referenced to the directions of the attached drawings. Therefore, the directional terms used are used to illustrate and not to limit the present invention.

[0027] Embodiment 1 Embodiment 1 provides an aircraft route mission control and execution system, see Figure 1 The steps include: Step 1: Construct a three-dimensional map model. The three-dimensional map model of the present invention is constructed based on Figure 2Taking a technical route in as an example, on the map engine Cesium (Cesium is a JavaScript library based on WebGL technology that can present high-precision three-dimensional earth and other celestial models on web pages for creating 3D earth and map applications. It not only supports multiple data formats (such as 3D Tiles, glTF, KML, etc.), but also has built-in rich interactive functions, such as address search, information window, full-screen viewing and WebVR support. The Cesium technology is an existing technology and is not described in detail in this application), a mesh model is placed on top of a 3D Gaussian splatting model for equal-proportion nesting to obtain a hybrid model of the mesh model and the 3D Gaussian splatting model, wherein the 3D Gaussian splatting model can process images to generate images that are closer to real scenes, and the grid-like images generated by the mesh model can add coordinate information, so the images generated by the hybrid model are real-life three-dimensional maps with coordinate information. The mesh model placement and 3D Gaussian splatting model are jointly driven by the 3D map engine. The nesting process is that the b3dm format file of the mesh model and the 3D Gaussian splatting model are jointly driven by the WebGL (WebGL is a JavaScript API for rendering 2D and 3D graphics in web browsers) technology on the map engine Cesium and the 3D engines Babylon.js and three.js. The 3D engines Babylon.js and three.js are integrated with the map engine Cesium. The integration of the 3D engines Babylon.js and three.js with the map engine Cesium uses map dynamic perspective binding and driving, direct binding association between map models and map engine parameters, plane cutting of 3D Gaussian splatting models using Cesium's own tools, and texture fusion of 3D Gaussian splatting models with Cesium screens. Because Cesium also has the characteristics of high performance and high precision, by optimizing the use of WebGL and making full use of hardware acceleration functions, efficient loading and rendering of large-scale 3D models and terrain data are achieved.

[0028] The captured images are brought into the hybrid model for training, and the images with coordinate information generated by the mesh model in the hybrid model are made transparent, and only the three-dimensional grid graphics with coordinate information are retained. The three-dimensional grid graphics are overlaid on the images generated by the 3D Gaussian splatting model training to obtain a three-dimensional map with three-dimensional coordinate data; Step 2: Select several points on the three-dimensional map to set as waypoints. Waypoints can be selected by directly clicking on the map. Since each point on the three-dimensional map obtained in step 1 has coordinate parameters, there is no need to recalculate the altitude when selecting a waypoint. Waypoints can be selected by clicking the mouse on the system or by touching the display screen of the system. After selecting the waypoints, the corresponding coordinate position information of the waypoints is recorded, and several waypoints are connected to form a route; Step 3: Assign functions to waypoints; There are two ways to assign functions to waypoints in step 3, namely, real scene function point selection and simulated function point selection. Real scene function point selection mainly includes the following steps: S3.1: Real-time video streaming, showing the images captured by the current camera of the aircraft; S3.2: For each waypoint selected in step 2 in turn, based on the image captured by the current camera and in accordance with the user's visual requirements, determine whether the shooting position and angle of the aircraft at the current waypoint meet the user's visual requirements. When the shooting position and angle of the aircraft do not meet the requirements, adjust the shooting position and angle of the aircraft, and adjust the three-dimensional coordinates of the current waypoint; the method for adjusting the three-dimensional coordinates of the current waypoint is to issue an adjustment command to the aircraft through the controller, and the aircraft adjusts the corresponding flight posture to obtain a shooting position and shooting angle that meet the requirements. The position of the aircraft after the posture is adjusted is the corrected three-dimensional coordinates of the waypoint.

[0029] The controller issues adjustment commands to the aircraft by sending flight instructions to the aircraft through the keyboard. For example, a more common method includes using the four keys W, S, A, and D on the keyboard (W stands for forward, S stands for backward, A stands for moving left, and D stands for moving right) to send instructions for controlling flight adjustments. The instructions are sent to the airport, and the airport receives the instructions to control the aircraft to execute the flight instructions, thereby achieving control and adjustment of the aircraft's flight.

[0030] In order to facilitate the flight control and adjustment of the aircraft, the voice control function can also be used. The user inputs voice into the system's voice recognition program, and the recognition program parses the voice to generate corresponding flight instructions and send them to the airport. The airport then receives the instructions to control the aircraft to execute the flight instructions, thereby achieving control and adjustment of the aircraft's flight.

[0031] S3.3: Assign corresponding functions to each adjusted waypoint. The functions assigned to the waypoints include: taking photos, changing the aircraft's yaw angle and setting the yaw angle value, changing the gimbal pitch angle and setting the gimbal pitch angle value, camera zoom and setting the zoom value, camera focus, switching lenses, taking photos, starting video recording, and ending video recording. After the aircraft flies to the waypoint, it will execute these functions according to the functions set at the waypoint to achieve the navigation effect.

[0032] The simulated function point selection is to assign waypoint functions to the waypoints set in step 2 in sequence, and the assigned waypoint functions are the same as those in step S3.3. Compared with the real scene point selection function, the simulated point selection function cannot adaptively modify the waypoint position according to the user's visual requirements.

[0033] Step 4: According to the waypoints and functions selected in Steps 2 and 3, simulate the flight, simulate the aircraft flying to each waypoint, and execute the flight effect of each waypoint. Compare the simulated flight effect with the actual flight effect, verify and modify the waypoints and waypoint functions in Step 3; When simulating the flight effect, the height and angle of the aircraft are calculated according to the position information of each point of the aircraft on the three-dimensional map, and the flight following effect is achieved in the first perspective. The first perspective following is achieved by simulation software. In this application, since the map will have coordinate information, it is easier to achieve the first perspective following.

[0034] Step 5: The aircraft receives the flight command, flies to the waypoint that meets the flight effect confirmed in step 4, and executes the corresponding waypoint function.

[0035] Among them, a GIS geographic information system is set up in the system, which calls the three-dimensional map in the Cesium map engine and realizes the simulation flight function, dynamic perspective calculation function and waypoint route marking function respectively.

[0036] In order to solve the problem that conventional drone mission management platforms can only fly to waypoints and execute waypoint functions, this application combines the aircraft route mission control and execution system with artificial intelligence visual algorithms to develop an application model that can be used to solve actual scene needs. The application models here include models for discovering real-life defects at waypoints, models for traffic congestion at waypoints, models for real-time monitoring of crowd flow at waypoints, etc. The specific implementation method is as follows: First, in the real-view point selection mode, the aircraft flies to the adjusted waypoint position multiple times and takes photos; the photos taken multiple times are sent to the artificial intelligence visual algorithm for training and the application model is constructed (the artificial visual intelligent algorithm is an existing technology application, which extracts and analyzes the input pictures to obtain the required application model); after receiving the flight command, the aircraft flies to the waypoint to take photos, and uses the constructed application model to compare and analyze the photos taken at the waypoint, find out the abnormal points in the pictures and output the decision plan. Taking the repair of glass curtain wall damage at the waypoint as an example, multiple photos B are taken at waypoint A through the real-view point selection function, and photos B are sent to the artificial intelligence visual algorithm for training to obtain an algorithm model for curtain wall defect detection. Through the monthly inspection of the aircraft, photos are taken at waypoint A, sent to the curtain wall defect detection algorithm model for detection, and the inspection results are output. Users can make corresponding decisions (whether to repair) based on the inspection results.

[0037] Due to the particularity of aerial photography, the pictures taken are limited and specific, so data enhancement is needed to improve the generalization ability of the model. In traditional solutions, image enhancement increases the diversity of training samples by rotation, scaling, flipping, translation, color jittering, etc. In order to enhance the data of the taken pictures, this application uses a generative adversarial network (GAN) to generate data samples to increase the diversity of samples. The specific steps include: S1: Generative adversarial networks include generators and discriminators. Generative adversarial networks randomly sample a vector from a latent space, which is often a noise vector that follows a uniform or normal distribution. This vector is used as the input of the generator, and an image data sample is gradually generated through the U-NET convolutional neural network. The generated sample and the real sample are sent to the discriminator network for discrimination. S2: The generator tries to generate more realistic samples to deceive the discriminator, while the discriminator tries to improve its discriminative ability to distinguish between real samples and generated samples. The parameters of the generator and discriminator are updated through the Adam back-propagation and optimization algorithm. The generator adjusts its own parameters by minimizing the output error of the discriminator to improve the fidelity of the generated samples. The discriminator adjusts its own parameters by minimizing the error of correctly classifying real and generated samples to improve its discriminative ability.

[0038] S3: Repeat steps S1 and S2 until the generator generates sufficiently realistic samples so that the discriminator cannot accurately distinguish the generated samples from the real samples. At this point, it can be considered that the generator has learned the distribution of real data and can generate samples similar to it. The generated samples are put into the data set and sent to the artificial intelligence vision algorithm for training.

[0039] At the same time, in order to improve the feature extraction capability of artificial intelligence vision algorithms (commonly referred to as computer vision (CV), which refers to a series of technologies that enable computers and machines to understand and interpret visual information. The goal of computer vision is to enable machines to extract useful information from images or videos and make decisions or take actions based on this information. This field combines deep learning, pattern recognition, image processing and other technologies to achieve various tasks) to improve the detection performance of image anomalies, this application makes corresponding modifications to the artificial intelligence vision algorithm. This application detects real scenes of aerial photography, which belongs to object detection. Commonly available visual algorithms include: YOLO (You Only Look Once): a fast object detection system that can predict multiple bounding boxes and their corresponding categories in a single network evaluation.

[0040] SSD (Single Shot MultiBox Detector): Similar to YOLO, SSD is also designed to achieve real-time object detection.

[0041] Faster R-CNN (Region-based Convolutional Neural Networks): Generates candidate regions through a region proposal network (RPN) and then uses a convolutional neural network for classification.

[0042] This application takes YOLO as an example. The backbone network of YOLO is a lightweight convolutional coding network. Image features are extracted by embedding a hierarchical attention network (EHAN). The feature fusion network uses a two-space aggregation pyramid (TSAP) for multi-scale feature fusion to adapt to targets of different sizes. The prediction layer uses a transpose encoder (Transpose Encoder) as the prediction layer to improve detection performance.

[0043] The input of the feature extraction network is an aerial image. First, the features are concentrated to the channel dimension through slicing operation, and then the image features are extracted through ordinary two-dimensional convolution and convolutional coding networks. Then, deeper features are extracted through the feature fusion network and spliced ​​with shallow features. The extracted features are then sampled to restore the original size of the image. Finally, four prediction layers are used to realize the detection of four levels: small targets, medium targets, large targets, and super-large targets.

[0044] In view of the particularity of aerial image imaging, this application uses a dynamic convolutional encoder (DCE) module for feature extraction in the embedded hierarchical attention network part, using small convolution kernels in the shallow network and large convolution kernels in the deep network, while ensuring a large receptive field and reducing computational complexity. The DCE module is combined with a transpose encoder (TE) to enhance the network's feature extraction capabilities, effectively integrating local information and global information to improve detection performance.

[0045] In the prior art, artificial intelligence vision algorithms use loss functions to optimize model parameters, and the loss function design can handle bounding box prediction and category classification at the same time. In view of the problem that the number of categories in the aerial photography detection dataset is unbalanced, resulting in an unbalanced training process, the traditional focal loss (Focal Loss) cannot effectively solve the problem of unbalanced scene categories. Therefore, this application combines the Complete Intersection over Union (CIoU) loss, the Binary Cross Entropy (BCE) loss, and the Equalized Focal Loss (EFL) to construct a new loss function that is conducive to balanced training.

[0046] For the regression mechanism of bounding boxes, there are two problems with the original IoU (Intersection over Union is a commonly used metric used to evaluate the degree of overlap between the predicted bounding box and the true bounding box in target detection tasks. It is very common in fields such as object detection and image segmentation and is used to quantify the similarity between two regions). First, when the true box and the predicted box do not overlap at all, the loss is a constant of 1, so it cannot reflect the distance between the predicted box and the true box; second, when the true box contains the predicted box and the sizes of the true box and the predicted box are fixed, no matter where the predicted box is in the true box, the loss is unchanged. CIoU (Complete Intersection over Union is an improved IoU measurement method that aims to solve the problem that traditional IoU cannot effectively provide gradient information in some cases. CIoU not only considers the overlapping area of ​​the two bounding boxes, but also the distance between the bounding boxes, the aspect ratio and other factors, thereby providing a more comprehensive and effective loss function in the training process) introduces a minimum bounding box c that contains the real box and the predicted box, and makes the center point of the predicted box close to the center point of the real box according to the straight-line distance between the predicted box and the center point of the bounding box. At the same time, CIoU also limits the aspect ratio of the predicted box to the real box, and uses the weight factor By balancing the IoU partial loss and the aspect ratio loss, CIoU can achieve better positioning effect through multi-dimensional loss calculation, and the training is more stable and the convergence speed is faster.

[0047] Focal Loss is a loss redistribution solution proposed mainly for the foreground-background imbalance problem. It focuses on the learning of foreground samples and reduces the attention paid to background interference information. This method is suitable for the situation where the foreground category is evenly distributed. However, the drone aerial photography dataset not only has the foreground-background imbalance problem, but also the foreground-foreground imbalance problem. Therefore, this method has great limitations. The balanced focus loss function of this application is based on Focal Loss. By introducing a focus factor and a weight factor, it can solve both the foreground-background imbalance and the foreground-foreground imbalance problems.

[0048] Confidence is used to indicate the credibility of the prediction box. In actual detection, it is necessary to first determine whether the prediction confidence of each prediction box exceeds the threshold. If it exceeds, it is determined that there is an object in the prediction box, and then the object classification and bounding box regression are performed. The confidence loss in this paper still adopts the binary cross entropy loss, which is combined with the fused classification and positioning loss and used as the total loss function of this paper.

[0049] In the formula, N represents the number of detection layers, represents the classification loss, represents the bounding box regression loss, is the confidence loss, , , Represents the weights assigned to the three losses.

[0050] The above is a detailed introduction to an aircraft route mission control and execution system provided by the present invention. The structure and working principle of the present invention are described in detail using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, the present invention can also be improved and modified without departing from the principle of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. An aircraft route mission control and execution system, characterized in that: The steps include: Step 1: Build a 3D map model. On a 3D map engine, place a mesh model on top of a 3D Gaussiansplatting model for nesting, so as to obtain a hybrid model of a mesh model and a 3D Gaussian splatting model. The 3D map engine jointly drives the placement of the mesh model and the 3D Gaussian splatting model, and brings the captured images into the hybrid model for training, so as to obtain a 3D map with 3D coordinate data required for flight routes. Step 2: Selecting a number of points on the three-dimensional map to set as waypoints, and recording the corresponding coordinate position information of the waypoints, and connecting a number of the waypoints to form a route; Step 3: Assign functions to waypoints; Step 4: According to the waypoints and waypoint functions selected in Steps 2 and 3, simulate the flight, simulate the aircraft flying to each waypoint, and execute the flight effect of each waypoint. Compare the simulated flight effect with the actual flight effect, and verify and modify the waypoints and waypoint functions in Step 3; Step 5: The aircraft receives the flight command, flies to the waypoint that meets the flight effect confirmed in step 4, and executes the corresponding waypoint function.

2. The aircraft route mission control and execution system according to claim 1, characterized in that: In the step 1, the mesh model and the 3D Gaussian splatting model are nested in equal proportions.

3. The aircraft route mission control and execution system according to claim 1, characterized in that: The image with coordinate information generated by the mesh model in the hybrid model in step 1 is made transparent, and only the three-dimensional grid graphics with coordinate information are retained. The three-dimensional grid graphics are overlaid on the image generated by the 3D Gaussian splatting model training to obtain a three-dimensional map with three-dimensional coordinate data.

4. The aircraft route mission control and execution system according to claim 1, characterized in that: Waypoint functions include: taking photos, changing the aircraft's yaw angle and setting the yaw angle value, changing the gimbal pitch angle and setting the gimbal pitch angle value, camera zoom and setting the zoom value, camera focus, switching lenses, taking photos, starting recording, and ending recording.

5. The aircraft route mission control and execution system according to claim 1, characterized in that: The waypoints are assigned functions by selecting points for real scene functions, which mainly includes the following steps: S3.1: Real-time video streaming, showing the images captured by the current camera of the aircraft; S3.2: For each waypoint selected in step 2, based on the image captured by the current camera and the visual requirements of the user, determine whether the shooting position and angle of the aircraft at the current waypoint meet the user's visual requirements. If the shooting position and angle of the aircraft do not meet the requirements, adjust the shooting position and angle of the aircraft, and adjust the three-dimensional coordinates of the current waypoint. S3.3: Assign corresponding functions to each adjusted waypoint.

6. The aircraft route mission control and execution system according to claim 5, characterized in that: In step S3.2, the method for adjusting the three-dimensional coordinates of the current waypoint is to send an adjustment command to the aircraft through the controller. The aircraft adjusts the corresponding flight posture to obtain a shooting position and shooting angle that meet the requirements. The position after the aircraft posture is adjusted is the corrected three-dimensional coordinates of the waypoint.

7. The aircraft route mission control and execution system according to claim 5, characterized in that: The control instructions received by the aircraft are sent to the airport based on the instructions issued by the keyboard or recognized by the voice recognition system. After the airport receives the instructions, it controls the aircraft to execute the corresponding instructions.

8. The aircraft route mission control and execution system according to claim 1, characterized in that: When simulating the flight effect, step 4 calculates the height and angle of the aircraft's flight based on the position information of each point of the aircraft on the three-dimensional map, and realizes the flight following effect from a first-person perspective.

9. The aircraft route mission control and execution system according to claim 1, characterized in that: In step 3, the waypoints are assigned functions in a simulated function point selection manner, and corresponding functions are assigned to each waypoint in turn based on the waypoints selected in step 2.

10. An aircraft route mission control and execution system according to any one of claims 1 to 8, characterized in that: The aircraft route mission control and execution system also includes, in the real-view function point selection mode, the aircraft flies to the adjusted waypoint position multiple times and takes photos; The photos taken multiple times are sent to the artificial intelligence visual algorithm for training to build the application model; After receiving the flight command, the aircraft flies to the waypoint to take photos, and uses the constructed application model to compare and analyze the photos taken at the waypoint to find abnormal points in the picture and output a decision plan.

11. The aircraft route mission control and execution system according to claim 10, characterized in that: Building an application model also includes: Dataset preparation: Use a generative adversarial network to enhance the data of the pictures taken by the aircraft. The generator of the generative adversarial network randomly samples a vector from a latent space as the input of the generator. The generator network receives the latent vector as input and gradually generates an image data sample through the U-NET convolutional neural network. The generated sample and the real sample are sent to the discriminator network for discrimination. The generator continuously generates data samples and sends them to the discriminator for identification until the generator generates sufficiently realistic samples so that the discriminator cannot accurately distinguish the generated samples from the real samples. At this time, the generated samples are put into the data set and sent to the artificial intelligence vision algorithm for training.

12. The aircraft route mission control and execution system according to claim 10, characterized in that: The lightweight convolutional coding network, the backbone network of the artificial intelligence vision algorithm, is improved. A hierarchical attention network is embedded in the lightweight convolutional coding network to extract image features in depth. The hierarchical attention network adopts a dynamic convolutional coding module for feature extraction, and uses small convolution kernels in shallow networks and large convolution kernels in deep networks.

13. The aircraft route mission control and execution system according to claim 10, characterized in that: The loss function used by the artificial intelligence vision algorithm for balance training is: ； In the formula, N represents the number of detection layers, represents the classification loss, represents the bounding box regression loss, is the confidence loss, , , Represents the weights assigned to the three losses.

Images