An aircraft visual navigation method based on double matching and dynamic satellite map loading

Abstract

The application discloses a kind of aircraft visual navigation methods based on double matching and dynamic satellite map loading, first offline preparation satellite slice database is used for subsequent online matching when dynamic loading;When visual matching positioning is carried out, the current frame is extracted and matched with satellite map according to the dynamic splicing loading of heading angle and height information, and the latitude and longitude positioning information P1 is obtained after solving;At the same time, the current frame is extracted and matched with the same feature of the previous frame, and the latitude and longitude positioning information P2 is obtained after solving;Finally, the double matching positioning result is input into the Bayesian fusion module for fusion optimization, and the final positioning result of the current frame is obtained.The method effectively solves the problem that traditional single matching method is prone to unstable and ineffective positioning, greatly improves the accuracy and robustness of matching positioning.

Description

Technical Field

[0001] This invention belongs to the field of aircraft technology, specifically relating to an aircraft visual navigation method based on dual matching and dynamic satellite image loading. Background Technology

[0002] High-precision positioning and navigation are key technologies for enabling autonomous flight of aircraft such as drones. Currently, common navigation schemes mainly rely on Global Positioning Systems (GPS) (such as the US GPS and China's BeiDou), which are widely used due to their high precision, all-weather capability, and flexibility. However, these navigation methods heavily depend on satellite signals, resulting in limitations such as weak autonomy, limited resistance to electromagnetic interference, and inability to be used in densely built-up areas or indoor environments. In contrast, positioning and navigation technology based on visual image matching, with its strong resistance to electromagnetic interference, low cost, and good environmental adaptability, has become an important way to improve the autonomous and reliable flight capabilities of aircraft in complex and dynamic environments. Its application potential is particularly significant when satellite signals are blocked due to interference.

[0003] Despite the extensive research and attention given to visual image matching navigation technology, existing methods still face the following challenges. First, the differences between heterogeneous images, as well as variations in viewpoint and scale, significantly affect the accuracy and robustness of the matching. Current methods (such as patents CN 117974791 A and CN 114199250 B) primarily utilize aerial and satellite images for matching and positioning. This single matching method is sensitive to environmental changes and is prone to problems such as matching instability or feature loss during the matching process. Second, traditional methods consume enormous computational resources when processing large-size satellite images, making it difficult for matching algorithms to meet the real-time requirements of edge computing platforms in aircraft.

[0004] In conclusion, to overcome the shortcomings of existing technologies, there is an urgent need to develop a more accurate and robust visual scene matching positioning and navigation technology. Simultaneously, this technology should be efficient and lightweight to optimize its application in actual aircraft positioning and navigation tasks. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides an aircraft visual navigation method based on dual matching and dynamic satellite image loading. First, a satellite slice database is prepared offline for dynamic loading during subsequent online matching. During visual matching and positioning, the current frame is matched with a dynamically stitched satellite image based on heading angle and altitude information, yielding latitude and longitude positioning information P1. Simultaneously, the current frame undergoes the same feature extraction and matching process with the previous frame, yielding latitude and longitude positioning information P2. Finally, the dual-matching positioning results are input into a Bayesian fusion module for optimization, resulting in the final positioning result for the current frame. This invention effectively solves the problem of unstable and failed positioning caused by traditional single-matching methods, significantly improving the accuracy and robustness of matching and positioning.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows:

[0007] Step 1: Preparation and preprocessing of satellite image slice database;

[0008] Download and prepare raw satellite images with real latitude and longitude location information, split them into several slices, prepare a satellite image slice image database, and store it on the edge computing platform of the spacecraft.

[0009] On the edge computing platform, a deep learning-based feature extraction and matching network model is pre-deployed to prepare for subsequent image processing and matching.

[0010] Step 2: System initialization;

[0011] The visual matching positioning system is initialized using the aircraft's initial position and heading angle information to obtain the initial visual matching area, thereby initiating the visual matching navigation process;

[0012] Step 3: Image input and dynamic satellite image loading;

[0013] The input end receives real-time image data captured by the downward-facing camera under the aircraft body to obtain the current frame image; at the same time, based on the positioning information of the previous frame, it uses a satellite image tiling and reloading strategy to load the corresponding regional satellite image to prepare for the next step of image matching.

[0014] Step 4: Dual matching and positioning;

[0015] Double matching is performed using a trained deep learning feature extraction and matching network model.

[0016] 1) Match and locate the aerial image of the current frame with the loaded satellite area image;

[0017] 2) Match and locate the aerial image of the current frame with the aerial image of the previous frame;

[0018] Step 5: Merge localization results;

[0019] The positioning information obtained from the double matching is integrated using a Bayesian fusion module to generate an optimized positioning result for the current frame.

[0020] Step 6: Save and iterate the results;

[0021] Store the current frame image and its location result, and use this location information as a constraint for matching and positioning in the next frame;

[0022] Return to step 3 to perform matching and positioning for the next frame;

[0023] This cycle continues, ensuring continuous positioning and navigation throughout the entire flight.

[0024] Preferably, step 1 specifically comprises:

[0025] First, the original satellite image is downloaded locally and split into N*N satellite sub-image slices, each slice being much smaller than the original satellite image. Each slice contains accurate latitude and longitude coordinates for subsequent dynamic stitching and recombination. Finally, the prepared slice image database is stored in the spacecraft edge computing platform. This process is completed offline before the matching algorithm is deployed.

[0026] Preferably, step 3 specifically comprises:

[0027] Based on the position and attitude information of the previous frame, a total of 9 slice images adjacent to the position of the previous frame are selected from the satellite slice database, and a continuous satellite region image is formed by image stitching technology. Then, using the heading angle and flight altitude information, the stitched satellite image is subjected to corresponding geometric transformations, including rotation, scaling and cropping.

[0028] Assuming the stitched satellite region image is Is, with width and height w*h, and the camera's horizontal and vertical field of view are FOV respectively. h and FOV v If the aircraft's flight altitude is H and its heading angle is θ, then the scaling and rotation process for the stitched image is as follows:

[0029] First, calculate the ground coverage width and height of the aerial images generated by the drone camera;

[0030]

[0031] Let the resolution of the satellite image be R. s Use equation (2) to calculate the scaling factor required for the satellite image:

[0032]

[0033] The satellite image is scaled using the scaling factor Scale_factor, and the adjusted image size is:

[0034]

[0035] Next, the scaled satellite image is rotated using the rotation matrix shown in equation (4), and a rectangular region is cropped to obtain the satellite image to be matched:

[0036]

[0037] Preferably, step 4 specifically comprises:

[0038] The XFeat network model based on deep learning is used to perform feature extraction and matching operations. In this process, features of the aerial image of the current frame, the generated satellite area image, and the aerial image of the previous frame are extracted to realize a dual matching mechanism of "sub-image-parent image" and "sub-image-sub-image".

[0039] The feature extraction and matching process is as follows:

[0040] 1) Feature extraction and matching;

[0041] For aerial image I1 and satellite image I2, feature extraction was performed using the XFeat feature extractor to obtain the extracted feature descriptors:

[0042]

[0043] After obtaining the descriptors, the features between the two images are matched to find the correspondence. This process is completed by the XFeat feature matcher, as shown in Equation (6):

[0044] M1_pts, M2_pts, conf=XFeat.match(Descriptors1,Descriptors2) (6)

[0045] 2) Solve for the homography transformation matrix and calculate the latitude and longitude coordinates;

[0046] The robust estimation algorithm MAGSAC is used to find the homography matrix H in the matching pair and to find the transformation relationship between the two images, as shown in Equation (7):

[0047] H= findHomography(M1_pts, M2_pts, MAGSAC) (7)

[0048] Then, using the obtained homography matrix, the center coordinates (P) of the UAV image are determined. cx ,Pcy Convert to pixel coordinates in satellite image (P') cx ,P' cy As shown in equation (4):

[0049] P c ′ x ,P c ′ y =PerspectiveTransform(P cx ,P cy ,H) (8)

[0050] Finally, based on the known latitude and longitude range (longitude [Lon1:Lon2], latitude [Lat1:Lat2]) and pixel size (w2*h2) of the satellite image, the actual latitude and longitude are calculated using equation (9):

[0051]

[0052] Perform matching and positioning operations on aerial image I1 and satellite image I2, as well as aerial image I1 and the previous frame aerial image I0, to obtain the latitude and longitude coordinate results and confidence scores of the double matching, namely the satellite image positioning result (lon1,lat1,conf1) and the inter-frame positioning result (lon2,lat2,conf2).

[0053] Preferably, step 5 specifically comprises:

[0054] A fusion module based on Bayesian theory is introduced to fuse the results of double matching.

[0055] The matching localization result between the current frame and the satellite image is denoted as P1(lon1,lat1,conf1); the inter-frame localization result is denoted as P2(lon2,lat2,conf2). Assuming that they follow independent probability distributions, the two sets of localization results are integrated through a Bayesian theory update mechanism to optimize the final localization output.

[0056] Satellite image matching and location distribution:

[0057]

[0058] Inter-frame matching and localization distribution:

[0059]

[0060] According to Bayesian theory, the fused localization result is obtained through equation (12):

[0061]

[0062] The confidence level after fusion is estimated using equation (13):

[0063]

[0064] A computer program that causes a computer to perform the aforementioned aircraft visual navigation method.

[0065] An electronic device includes a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to enable the electronic device to perform the above-described aircraft visual navigation method.

[0066] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned aircraft visual navigation method.

[0067] A chip includes a processor for retrieving and running a computer program from a memory, causing a device equipped with the chip to perform the aforementioned aircraft visual navigation method.

[0068] A computer program product includes a computer storage medium storing a computer program, the computer program including instructions executable by at least one processor, which, when executed by the at least one processor, implement the above-described aircraft visual navigation method.

[0069] The beneficial effects of this invention are as follows:

[0070] 1) The dual-matching method in this invention improves positioning accuracy and robustness in complex environments: This invention employs a dual-matching method, including matching the current frame with satellite imagery and matching the current frame with the previous frame, and integrates the results through a Bayesian fusion module. This method effectively solves the problem that traditional single-matching methods are prone to positioning instability and failure, significantly improving the accuracy and robustness of matching positioning.

[0071] 2) The dynamic reassembly and loading method in this invention reduces the computational resource consumption of the visual matching localization algorithm: by dividing the original satellite image into smaller slices and dynamically loading these small-sized satellite sub-images during the matching process, this invention significantly reduces the consumption of computational resources. This method effectively solves the problems of excessive computational burden and memory consumption in traditional methods. While improving the real-time performance of the algorithm, it reduces the matching search space, thereby further improving the localization accuracy.

[0072] 3) This invention provides a highly efficient and adaptable visual matching positioning method, achieving excellent performance on resource-constrained platforms. This method is suitable for long-distance positioning and navigation tasks in complex environments such as spacecraft satellite denial, demonstrating promising application prospects and significant engineering practical value. Attached Figure Description

[0073] Figure 1 This is a diagram illustrating the implementation steps of the present invention;

[0074] Figure 2 This is a framework for a visual positioning and navigation method based on dual matching and dynamic satellite image loading.

[0075] Figure 3 This is a schematic diagram of satellite image tiling and dynamic loading process;

[0076] Figure 4 This is a schematic diagram of a dual-matching positioning mechanism;

[0077] Figure 5 This is a schematic diagram of the matching and localization process using the XFeat neural network model;

[0078] Figure 6 Flowchart for Bayesian fusion module processing;

[0079] Figure 7 This is an experimental implementation device for the visual matching positioning and navigation system according to an embodiment of the present invention;

[0080] Figure 8 This is a diagram illustrating the flight path of the UAV during an experiment according to an embodiment of the present invention;

[0081] Figure 9 This is a double-matching result from an embodiment of the present invention;

[0082] Figure 10 The results are from the test implementation of the visual matching positioning navigation of this invention. Detailed Implementation

[0083] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0084] This invention aims to address the problems of decreased accuracy and insufficient robustness caused by differences in image sources and variations in viewpoint and scale in existing single-vision scene matching methods, as well as the challenge of poor real-time performance on resource-constrained edge platforms due to excessive computational resource consumption. To this end, this invention proposes an aircraft positioning and navigation method and apparatus based on a dual matching strategy and dynamic satellite master image tile reconstruction loading. This innovative scheme can significantly improve the accuracy and robustness of matching positioning and achieve excellent real-time performance on edge computing platforms.

[0085] This invention proposes a UAV positioning and navigation method based on a dual-matching method (matching aerial sub-images with satellite master images and inter-frame sub-image matching) and a satellite image tile stitching and loading strategy. The implementation steps are as follows: Figure 1 As shown:

[0086] Step 1: Preparation and preprocessing of satellite baseline images;

[0087] To achieve efficient and accurate matching, this invention first downloads and prepares a satellite reference image with real latitude and longitude location information, and stores it on the spacecraft's edge computing platform. To optimize computing resources and accelerate the matching process, the original satellite image is split into several small-sized slices to adaptively load the regions to be processed. On the edge computing platform, a deep learning-based feature extraction and matching network model is pre-deployed to prepare for subsequent efficient image processing and matching.

[0088] Step 2: System initialization;

[0089] The visual matching positioning system is initialized using the aircraft's initial position and heading angle information to obtain the initial visual matching area, thereby initiating the visual matching navigation process. This step ensures that the system can quickly enter a stable working state.

[0090] Step 3: Image Input and Dynamic Satellite Image Loading;

[0091] The algorithm's input receives real-time image data captured by a downward-facing camera beneath the aircraft, obtaining the current frame image. Simultaneously, based on the positioning information from the previous frame (or the initial frame), the satellite image tiling and reloading strategy designed in this invention is applied to load the corresponding small-area satellite image, preparing for the next step of image matching.

[0092] Step 4: Dual matching and positioning;

[0093] A dual matching process is performed using a trained deep learning feature extraction and matching network model. This step includes:

[0094] 1) Match and locate the aerial image of the current frame with the loaded satellite area image.

[0095] 2) Match and locate the aerial image of the current frame with the aerial image of the previous frame.

[0096] Step 5: Merge the localization results;

[0097] The localization information obtained from the dual matching is integrated using a Bayesian fusion module to generate an optimized localization result for the current frame. This process, through confidence-based fusion optimization, significantly improves the accuracy and robustness of localization.

[0098] Step Six: Saving and Iterating Results;

[0099] The current frame image and its positioning result are stored, and this positional information is used as a constraint for matching and positioning in the next frame. Return to step three to perform matching and positioning for the next frame. This loop continues to ensure continuous and high-precision positioning and navigation throughout the entire flight.

[0100] The overall processing flow of this invention is as follows: Figure 2 As shown, a satellite slice database is first prepared offline for dynamic loading during subsequent online matching. During visual matching and localization, the current frame is matched with a dynamically stitched satellite image based on heading angle and altitude information, yielding latitude and longitude positioning information P1. Simultaneously, the current frame undergoes the same feature extraction and matching process with the previous frame, resulting in latitude and longitude positioning information P2. Then, the dual-matching localization results are input into a Bayesian fusion module for optimization, yielding the final localization result for the current frame.

[0101] The specific implementation schemes involved in the above technical framework are as follows:

[0102] (1) Preparation of satellite image slice image database

[0103] First, the original satellite imagery is downloaded locally and split into N*N sub-image tiles, each much smaller than the original image. Each tile contains accurate latitude and longitude coordinates, which can be used for subsequent dynamic stitching and reconstruction. Finally, the prepared tile image database is stored in the spacecraft's edge computing platform; this process is completed offline before the matching algorithm is deployed.

[0104] (2) Preprocessing and dynamic loading of satellite regional images

[0105] Based on the position and attitude information of the previous frame (or the initial frame), nine adjacent slices are selected from the satellite slice database and stitched together to form a continuous satellite region image. Next, to correct the perspective differences between the satellite image and the aerial photograph, geometric transformations, including rotation, scaling, and cropping, are performed on the stitched satellite image using heading angle and flight altitude information. This step aims to generate a reference image with a higher degree of matching to the current aerial photograph frame, thereby enhancing the accuracy of the image matching process.

[0106] Assuming the stitched satellite region image is Is, with width and height w*h, and the camera's horizontal and vertical field of view are FOV respectively. h and FOV v The aircraft's flight altitude is H, and its heading angle is θ. The scaling and rotation process for the stitched image is as follows:

[0107] 1) First, calculate the ground coverage width and height of the aerial images generated by the drone camera.

[0108]

[0109] 2) Let the resolution of the satellite image be Rs. Use equation (2) to calculate the scaling factor required for the satellite image. In order to avoid image distortion caused by scaling, choose to use a smaller scaling factor:

[0110]

[0111] 3) Scale the satellite image using the scaling factor (Scale_factor). The adjusted image size is:

[0112]

[0113] 4) Next, rotate the scaled satellite image using the rotation matrix shown in equation (4) and crop out the rectangular area to obtain the satellite image to be matched.

[0114]

[0115] Satellite image tile reconstruction and scaling / rotation loading process as follows Figure 3 As shown:

[0116] (3) Dual feature extraction and matching localization;

[0117] This invention utilizes the XFeat network model based on deep learning for feature extraction and matching. During this process, features are extracted from the aerial image (sub-image) of the current frame, the generated satellite region image (mother image), and the aerial image of the previous frame (sub-image), achieving a dual matching mechanism of "sub-image-mother image" and "sub-image-sub-image," such as... Figure 4 As shown in the image.

[0118] The XFeat model is a highly efficient neural network architecture that employs a multi-scale feature detection framework to identify and extract salient structures from local to global perspectives in images. This multi-scale feature detection method provides robustness for image processing under different viewpoints and zoom conditions. Because XFeat uses a lightweight CNN structure, it can run efficiently on computationally limited platforms while meeting the needs of applications requiring high throughput or high computational efficiency. Its matching refinement module can extract pixel-level offsets from coarse and semi-dense matching, achieving more accurate matching results. In this invention, the "double matching" step uses the same shared XFeat model to simplify the system architecture and improve processing consistency. Figure 5 As shown.

[0119] The feature extraction and matching process is as follows:

[0120] 1) Feature extraction and matching;

[0121] For aerial image I1 and satellite image I2, feature extraction was performed using the XFeat feature extractor to obtain the extracted feature descriptors:

[0122]

[0123] After obtaining the descriptors, the features between the two images are matched to find the correspondence. This process is completed by the XFeat feature matcher, as shown in Equation (6):

[0124] M1_pts, M2_pts, conf=XFeat.match(Descriptors1,Descriptors2) (19)

[0125] 2) Solve for the homography transformation matrix and calculate the latitude and longitude coordinates;

[0126] Next, the more advanced robust estimation algorithm MAGSAC (Maximum A Posteriori SAmpleConsensus) is used to find the homography matrix H in the matching pair in order to find the transformation relationship between the two images, as shown in Equation (7):

[0127] H= findHomography(M1_pts, M2_pts, MAGSAC) (20)

[0128] Then, using the obtained homography matrix, the center coordinates (P) of the UAV image are determined. cx ,P cy Convert to pixel coordinates in satellite image (P') cx ,P' cy As shown in equation (4):

[0129] P c ′ x ,P c ′ y =PerspectiveTransform(P cx ,P cy ,H) (21)

[0130] Finally, based on the known latitude and longitude range (longitude [Lon1:Lon2], latitude [Lat1:Lat2]) and pixel size (w2*h2) of the satellite image, the actual latitude and longitude are calculated using equation (9):

[0131]

[0132] By performing the above matching and positioning operations on aerial image I1 and satellite image I2, as well as aerial image I1 and the previous frame aerial image I0, we can obtain the latitude and longitude coordinate results and confidence scores of the double matching, namely the satellite image positioning result (lon1,lat1,conf1) and the inter-frame positioning result (lon2,lat2,conf2).

[0133] (4) Bayesian fusion module;

[0134] To improve the accuracy and robustness of positioning results and avoid visual positioning algorithm failure due to inaccurate or failed matching of aerial and satellite images, this invention introduces a Bayesian-based fusion module to fuse the dual matching results. The processing flow of this fusion module is as follows: Figure 6 As shown.

[0135] The Bayesian fusion algorithm processes the following: The matching localization result between the current frame and the satellite image is denoted as P1(lon1,lat1,conf1), and the inter-frame localization result is denoted as P2(lon2,lat2,conf2). It is assumed that they each follow independent probability distributions. Then, the two sets of localization results are integrated through a Bayesian theoretical update mechanism to optimize the final localization output.

[0136] Satellite image matching and location distribution:

[0137]

[0138] Inter-frame matching and localization distribution:

[0139]

[0140] According to Bayesian theory, the fused localization result is obtained through equation (12):

[0141]

[0142] The confidence level after fusion is estimated using equation (13):

[0143]

[0144] Example:

[0145] To verify the effectiveness of the UAV visual matching navigation method based on dual matching and dynamic slice loading proposed in this invention, it was deployed on the embedded edge computing platform NVIDIA Orin Nano, and flight experiments were conducted on a UAV platform. Specific implementation examples are as follows:

[0146] (1) Implementation device and its components;

[0147] The experimental setup used in this embodiment consists of an octagonal rotary-wing UAV, the NVIDIA OrinNano edge computing platform, and a visible light pod. Figure 7 As shown in the diagram, the rotary-wing UAV is equipped with a high-precision RTK positioning module to provide accurate baseline positioning values. The NVIDIA Orin Nano, as an edge computing platform, boasts 40 TOPS of deep learning computing power, a power consumption range of 7.5-15W, and deploys the visual matching positioning and navigation algorithm designed in this invention. A visible light pod connects to the NVIDIA Orin Nano, providing real-time downward-facing images to support the operation of the visual positioning algorithm. The UAV is also equipped with a data transmission module, capable of transmitting the visual matching positioning results to a ground monitoring station in real time for analysis.

[0148] (2) Experimental environment;

[0149] The experimental verification was conducted in a location with diverse terrain, including grasslands, farmland, playgrounds, and building complexes, to fully test the algorithm's accuracy and robustness. The drone flew along a preset route, as shown in the example flight path. Figure 8 As shown, the flight area's latitude and longitude range is: longitude [108.7451474°E, 108.7728831°E], latitude [34.04564629°N, 34.02483682°N]. The flight altitude is 500 meters, the average flight speed is 10.5 meters per second, and the total flight distance is approximately 24.5 kilometers.

[0150] (3) Analysis of experimental results;

[0151] Figure 9 The image shows the result of dual feature matching for a frame during flight. As can be seen, due to the difference in perspective, there are relatively few matching point pairs between the aerial image and the satellite image in this frame (left image). This situation easily leads to the failure of traditional single-matching algorithms. However, the inter-frame matching added in this invention still matches a sufficient number of feature point pairs (right image). Therefore, performing dual matching between the current frame and the satellite image, as well as with the previous aerial image, can effectively avoid the problem of single-matching failure in complex scenes and effectively enhance the robustness of the matching and localization method.

[0152] Figure 10 The experimental results shown illustrate the positioning trajectories of the visual matching flight test, where the yellow trajectory represents the actual flight path and the green trajectory represents the matching positioning path achieved through this invention. Qualitative analysis shows that the two trajectories highly overlap, indicating that this invention has excellent positioning accuracy. During the experiment, this invention maintained stable matching positioning functionality under various complex turning paths, without any matching loss, demonstrating its superior robustness. Figure 10Quantitative analysis of the real-time error output shows that the positioning errors of this invention in longitude and latitude are less than 2.5% and 3.0%, respectively, representing a significant improvement over existing technologies. In terms of real-time performance, the total positioning time per frame of this invention is less than 30ms, which can well meet the real-time requirements of airborne systems.

Claims

1. A visual navigation method for aircraft based on dual matching and dynamic satellite image loading, characterized in that, Includes the following steps: Step 1: Preparation and preprocessing of satellite image slice database; Download and prepare raw satellite images with real latitude and longitude location information, split them into several slices, prepare a satellite image slice image database, and store it on the edge computing platform of the spacecraft. On the edge computing platform, a deep learning-based feature extraction and matching network model is pre-deployed to prepare for subsequent image processing and matching. Step 2: System initialization; The visual matching positioning system is initialized using the aircraft's initial position and heading angle information to obtain the initial visual matching area, thereby initiating the visual matching navigation process; Step 3: Image input and dynamic satellite image loading; The input end receives real-time image data captured by the downward-facing camera under the aircraft body to obtain the current frame image; at the same time, based on the positioning information of the previous frame, it uses a satellite image tiling and reloading strategy to load the corresponding regional satellite image to prepare for the next step of image matching. Step 4: Dual matching and positioning; Double matching is performed using a trained deep learning feature extraction and matching network model. 1) Match and locate the aerial image of the current frame with the loaded satellite area image; 2) Match and locate the aerial image in the current frame with the aerial image in the previous frame; The XFeat network model based on deep learning is used to perform feature extraction and matching operations. In this process, features of the aerial image of the current frame, the generated satellite area image, and the aerial image of the previous frame are extracted to realize a dual matching mechanism of "sub-image-parent image" and "sub-image-sub-image". The feature extraction and matching process is as follows: 1) Feature extraction and matching; For aerial image I1 and satellite image I2, feature extraction was performed using the XFeat feature extractor to obtain the extracted feature descriptors: (5) After obtaining the descriptors, the features between the two images are matched to find the correspondence. This process is completed by the XFeat feature matcher, as shown in Equation (6): (6) 2) Solve for the homography transformation matrix and calculate the latitude and longitude coordinates; The robust estimation algorithm MAGSAC is used to find the homography matrix H1 in the matching pair and to find the transformation relationship between the two images, as shown in Equation (7): (7) Then, the obtained homography matrix is ​​used to determine the center coordinates of the UAV image. Pixel coordinates converted to satellite image As shown in equation (8): (8) Finally, based on the known latitude and longitude range of the satellite imagery: longitude [Lon1:Lon2], latitude [Lat1:Lat2], and pixel size... w 2 *h 2. Calculate the actual latitude and longitude using equation (9): (9) Perform matching and positioning operations on aerial image I1 and satellite image I2, as well as aerial image I1 and the previous frame aerial image I0, to obtain the latitude and longitude coordinate results and confidence scores of the double matching, namely the satellite image positioning result (lon1, lat1, conf1) and the inter-frame positioning result (lon2, lat2, conf2). Step 5: Merge localization results; The positioning information obtained from the double matching is integrated using a Bayesian fusion module to generate an optimized positioning result for the current frame. Step 6: Save and iterate the results; Store the current frame image and its location result, and use this location information as a constraint for matching and positioning in the next frame; Return to step 3 to perform matching and positioning for the next frame; Repeat steps 3 through 6 to ensure continuous positioning and navigation throughout the flight.

2. The aircraft visual navigation method based on dual matching and dynamic satellite image loading according to claim 1, characterized in that, Step 1 specifically involves: First, the original satellite image is downloaded locally and split into N*N satellite sub-image tiles, each tile being much smaller than the original satellite image. Each tile contains accurate latitude and longitude coordinates for subsequent dynamic stitching and recombination. Finally, the prepared tile image database is stored in the spacecraft edge computing platform. This step is completed offline before the matching algorithm is deployed.

3. The aircraft visual navigation method based on dual matching and dynamic satellite image loading according to claim 2, characterized in that, Step 3 specifically involves: Based on the position and attitude information of the previous frame, a total of 9 slice images adjacent to the position of the previous frame are selected from the satellite slice database, and a continuous satellite region image is formed by image stitching technology. Then, using the heading angle and flight altitude information, the stitched satellite image is subjected to corresponding geometric transformations, including rotation, scaling and cropping. Assuming the stitched satellite region image is Is, its width and height are... w*h The camera's horizontal and vertical field of view are FOV and FOV, respectively. h and FOV v The aircraft is flying at an altitude of H and a heading angle of H. θ The scaling and rotation process for the stitched image is as follows: First, calculate the ground coverage width and height of the aerial images generated by the drone camera: (1) Assume the resolution of the satellite image is R s Use equation (2) to calculate the scaling factor required for the satellite image: (2) The satellite image is scaled using the scaling factor Scale_factor, and the adjusted image size is: (3) Next, the scaled satellite image is rotated using the rotation matrix shown in equation (4), and a rectangular region is cropped to obtain the satellite image to be matched: (4)。 4. The aircraft visual navigation method based on dual matching and dynamic satellite image loading according to claim 3, characterized in that, Step 5 specifically involves: A fusion module based on Bayesian theory is introduced to fuse the results of double matching. The matching localization result between the current frame and the satellite image is denoted as P1(lon1,lat1,conf1); the inter-frame localization result is denoted as P2(lon2,lat2,conf2). Assuming that they follow independent probability distributions, the two sets of localization results are integrated through a Bayesian theory update mechanism to optimize the final localization output. Satellite image matching and location distribution: (10) Inter-frame matching and localization distribution: (11) According to Bayesian theory, the fused localization result is obtained through equation (12): (12) The confidence level after fusion is estimated using equation (13): (13)。 5. An electronic device, characterized in that, include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 4.

7. A chip, characterized in that, include: A processor for retrieving and running a computer program from memory, causing a device on which the chip is mounted to perform the method as described in any one of claims 1 to 4.

8. A computer program product, characterized in that, The computer program product includes a computer storage medium storing a computer program, the computer program including instructions executable by at least one processor, which, when executed by the at least one processor, implement the method as described in any one of claims 1 to 4.

Images