Method, system and device for generating flight simulation terrain by multispectral fusion and storage medium

Abstract

This application relates to the field of flight simulation technology, providing a method, system, device, and storage medium for multispectral fusion-based flight simulation terrain generation, addressing the problem of low accuracy in flight simulation terrain generation. The method includes: selecting target multispectral images with cloud coverage below a set threshold and target synthetic aperture radar data for the corresponding time window; extracting terrain edge features, dynamically adjusting the radar beam incident angle, and generating point cloud data with non-uniform resolution; constructing a dual-branch processing model, with the first branch processing the target multispectral images and the second branch processing the point cloud data, obtaining fusion features through a weighted mechanism; dynamically associating the vegetation index distribution of the terrain area to be measured with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result; and generating a three-dimensional flight simulation terrain model with semantic labels based on the spatiotemporal dynamic mapping result for flight simulation path planning. This application improves the accuracy of flight simulation terrain generation.

Description

Technical Field

[0001] This application relates to the field of flight simulation technology, and in particular to a method and system for generating flight simulation terrain through multispectral fusion. Background Technology

[0002] Flight simulation terrain refers to a virtual 3D terrain model constructed using technical means that matches the features of real terrain. It is mainly used in scenarios such as flight training, UAV path planning, and aviation mission simulation. These scenarios have extremely high requirements for the accuracy, detail richness, and dynamic adaptability of the terrain model. The model needs to accurately reflect features such as terrain undulation, vegetation cover, and crop types, while maintaining sufficient detail in complex terrain areas (such as steep slopes and canyons). At the same time, it needs to be able to cope with different weather conditions to ensure the realism and reliability of the flight simulation.

[0003] Currently, one approach to generating terrain for flight simulations is based on a single multispectral image data processing scheme. This scheme acquires multispectral images of the terrain area to be measured, directly performs band splitting and feature extraction on the multispectral images to generate terrain texture information, and then combines this with preset elevation data to construct a 3D terrain model. Specifically, it first selects image data without cloud cover, identifies the approximate terrain outline through the distribution patterns of pixel values ​​in the images, and then simply overlays the extracted texture information with elevation data at a fixed resolution to form a terrain model for flight simulation. The entire process does not incorporate other types of remote sensing data for supplementary optimization.

[0004] However, the existing solution has the following obvious drawbacks: First, it relies solely on multispectral imagery data, which is susceptible to weather conditions. When cloud cover is present, it can lead to inaccurate extraction of terrain texture information and make it impossible to effectively filter out high-quality imagery data. Second, the generated elevation data uses a uniform resolution, which results in insufficient detail representation in complex terrain areas (such as steep slopes and canyons), making it difficult to accurately reflect the true undulations of the terrain. Third, vegetation information is not well represented in the terrain model, making it difficult to take into account the impact of vegetation cover on flight safety when planning flight simulation paths. Summary of the Invention

[0005] This application provides a multispectral fusion method and system for generating flight simulation terrain, which solves the problems of low accuracy in flight simulation terrain generation caused by low image data quality, difficulty in accurately reflecting the real undulations of the terrain, and lack of vegetation information in the prior art.

[0006] In a first aspect, this application provides a multispectral fusion method for generating flight simulation terrain, including:

[0007] Collect raw multispectral images and raw synthetic aperture radar data of the terrain area to be measured. Identify cloud coverage based on the pixel value distribution of the raw multispectral images. Filter target multispectral images and target synthetic aperture radar data of the corresponding time window with cloud coverage below a set threshold.

[0008] Topographic edge features are extracted from the target multispectral image, and the radar beam incident angle is dynamically adjusted based on the topographic edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density of the complex terrain area is increased by increasing the tilt of the radar beam incident angle. The complex terrain area includes steep slopes and canyon areas.

[0009] A dual-branch processing model is constructed, with the first branch processing the target multispectral image and the second branch processing the point cloud data. The high-resolution terrain features output from the two branches are fused through a weighted mechanism to obtain the fused features.

[0010] The vegetation index distribution of the terrain area to be measured is dynamically correlated with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of vegetation cover area terrain and crop type.

[0011] Based on the spatiotemporal dynamic mapping results, a three-dimensional flight simulation terrain model with semantic labels is generated for flight simulation path planning.

[0012] Optionally, the step of dynamically adjusting the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution, and increasing the point cloud density in complex terrain areas by increasing the tilt of the radar beam incident angle during the generation process, includes:

[0013] By analyzing the terrain edge features, the outline of the terrain region to be measured is generated, the curvature change rate of the outline is calculated, and terrain edge feature lines are generated based on the curvature change rate.

[0014] Based on the spatial distribution pattern of the terrain edge feature lines, the terrain area to be measured is divided into multiple sub-regions with similar terrain features, and the corresponding dominant slope aspect of each sub-region is determined.

[0015] Based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region, the complexity value of each sub-region is calculated, and the radar beam incident angle is adjusted according to the complexity value.

[0016] The control radar scans each of the sub-regions according to the adjusted radar beam incident angle, obtains the reflection signal of each of the sub-regions, analyzes the return time and intensity of the reflection signal, and determines the position of the first reflection point.

[0017] An initial point cloud data set is generated based on the position of the first reflection point. In the initial point cloud data, for areas with complex terrain, the point cloud density of the complex terrain area is increased by increasing the tilt of the radar beam incident angle, so as to generate point cloud data with non-uniform resolution.

[0018] Optionally, calculating the complexity value of each sub-region based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region includes:

[0019] The total length of all the terrain edge feature lines in each of the sub-regions is calculated, and the total length is divided by the area of ​​the corresponding sub-region to obtain the density value of the terrain edge feature lines in the unit sub-region.

[0020] Analyze the range of the angle between the dominant slope aspect and the due north direction in each of the sub-regions, and divide the range of the angle into multiple angle intervals;

[0021] Determine the corresponding slope aspect coefficient based on the included angle range of the dominant slope aspect of each sub-region;

[0022] Calculate the average curvature of the terrain edge feature lines in each of the sub-regions, and divide the average curvature into multiple levels, with each level of the average curvature corresponding to a different complexity correction factor;

[0023] The product of the density value of the terrain edge feature line in the unit sub-region, the corresponding slope aspect coefficient, and the corresponding complexity correction factor is used as the complexity value of each sub-region.

[0024] Optionally, the step of generating an initial point cloud data set based on the position of the first reflection point, and increasing the point cloud density in the complex terrain area by increasing the tilt of the radar beam incident angle to generate point cloud data with non-uniform resolution, includes:

[0025] Arrange the position information of all the first reflection points according to three-dimensional spatial coordinates to generate an initial point cloud data set;

[0026] Each initial point cloud data in the initial point cloud dataset is assigned to a corresponding sub-region;

[0027] Extract the sub-regions that are considered to be complex terrain regions from all sub-regions that have been allocated initial point cloud data, and control the target synthetic aperture radar to scan the complex terrain regions multiple times according to the target radar beam incident angle tilt to obtain the location information of the second reflection point. The target radar beam incident angle tilt is greater than the adjusted radar beam incident angle.

[0028] The location information of the second reflection point is added to the initial point cloud data of the corresponding complex terrain area to obtain the target point cloud data of the complex terrain area, so as to improve the point cloud density of the complex terrain area.

[0029] By comparing the point cloud density corresponding to each of the sub-regions, it is verified that the point cloud density of the terrain-complex region is higher than that of the non-terrain-complex region in the terrain region to be tested. After the verification is passed, point cloud data with non-uniform resolution is generated based on the target point cloud data of the terrain-complex region and the initial point cloud data of the non-terrain-complex region.

[0030] Optionally, the high-resolution terrain features include texture information and geometric information;

[0031] The constructed dual-branch processing model involves the first branch processing the target multispectral image and the second branch processing the point cloud data. A weighted mechanism is used to fuse the high-resolution terrain features output from both branches to obtain fused features, including:

[0032] Based on the first branch, the target multispectral image is decomposed into single-band images corresponding to multiple bands, the spatial distribution pattern of pixel values ​​in each single-band image is determined, and texture information corresponding to terrain features is extracted based on the spatial distribution pattern.

[0033] Based on the second branch, the point cloud data is organized into multiple data groups according to spatial location relationships. The distribution density and relative location relationships of the point cloud data in each data group are analyzed, and geometric information corresponding to terrain features is extracted.

[0034] The confidence weight of the texture information is determined based on the acquisition parameters of the target multispectral image, and the confidence weight of the geometric information is determined based on the acquisition parameters of the point cloud data.

[0035] The texture information and the geometric information are weighted and merged according to their corresponding confidence weights to obtain the fused feature.

[0036] Optionally, the step of decomposing the target multispectral image into multiple single-band images corresponding to each band based on the first branch, determining the spatial distribution pattern of pixel values ​​in each single-band image, and extracting texture information corresponding to terrain features based on the spatial distribution pattern includes:

[0037] The target multispectral image is divided into multiple independent single-band images according to a preset light wavelength range;

[0038] Based on the values ​​of adjacent pixels in each single-band image, determine the positions where the value difference between adjacent pixels in each single-band image is greater than a preset difference threshold. According to the positions where the value difference is greater than the preset difference threshold, divide the corresponding single-band image and record the distribution characteristics of each region in the single-band image.

[0039] Based on the distribution characteristics of each region, determine the spatial distribution pattern of pixel values ​​in each single-band image;

[0040] Based on the shape, topographic relief, and topographic morphology of each region in the spatial distribution pattern, texture information corresponding to the topographic features is extracted from the single-band image.

[0041] Optionally, the step of dynamically associating the vegetation index distribution of the topographic region to be measured with the elevation information in the fused features to generate a spatiotemporal dynamic mapping result of vegetation cover topography and crop type includes:

[0042] Images of specific bands are selected from the target multispectral image, and the vegetation index distribution of the terrain area to be measured is obtained by calculating the proportional relationship of the pixel values ​​corresponding to vegetation in the single-band images of the specific bands.

[0043] Elevation information is extracted from the fused features, and the range of elevation values ​​in the terrain region to be measured is determined based on the elevation information;

[0044] The vegetation index distribution and the elevation information are correlated according to spatial location to obtain the degree of correlation between the vegetation index distribution value and the elevation value at each spatial location.

[0045] Based on the range of changes in the elevation values, the correlation degree is adjusted, and based on the adjusted correlation degree, combined with the growth patterns of crop types under the elevation information and the vegetation index distribution, a spatiotemporal dynamic mapping result is generated.

[0046] Secondly, this application provides a multispectral fusion flight simulation terrain generation system, comprising:

[0047] The acquisition module is used to acquire raw multispectral images and raw synthetic aperture radar data of the terrain area to be measured, identify cloud coverage based on the pixel value distribution of the raw multispectral images, and filter target multispectral images and target synthetic aperture radar data with cloud coverage below a set threshold and corresponding time windows.

[0048] The extraction module is used to extract terrain edge features from the target multispectral image and dynamically adjust the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density of the terrain complex area is increased by increasing the tilt of the radar beam incident angle. The terrain complex area includes steep slopes and canyon areas.

[0049] The module is used to build a dual-branch processing model. The first branch processes the target multispectral image, and the second branch processes the point cloud data. The high-resolution terrain features output by the two branches are fused through a weighted mechanism to obtain the fused features.

[0050] The association module is used to dynamically associate the vegetation index distribution of the terrain area to be measured with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of vegetation cover area terrain and crop type.

[0051] The generation module is used to generate a three-dimensional flight simulation terrain model with semantic tags based on the spatiotemporal dynamic mapping results, for use in flight simulation path planning.

[0052] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a multispectral fusion flight simulation terrain generation method as described in any of the first aspects.

[0053] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a multispectral fusion flight simulation terrain generation method as described in any of the first aspects.

[0054] This application provides a multispectral fusion-based method for generating flight simulation terrain, comprising: acquiring original multispectral images and original synthetic aperture radar (SAR) data of the terrain area to be measured; identifying cloud cover based on the pixel value distribution of the original multispectral images; filtering target multispectral images and corresponding time window target SAR data with cloud cover below a set threshold; extracting terrain edge features from the target multispectral images; dynamically adjusting the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution; increasing the tilt of the radar beam incident angle during generation to improve the point cloud density in complex terrain areas, including steep slopes and canyons; constructing a dual-branch processing model, with the first branch processing the target multispectral images and the second branch processing the point cloud data; fusing the high-resolution terrain features output by the two branches through a weighted mechanism to obtain fused features; dynamically associating the vegetation index distribution of the terrain area to be measured with the elevation information in the fused features to generate a spatiotemporal dynamic mapping result of vegetation cover terrain and crop type; and generating a three-dimensional flight simulation terrain model with semantic labels based on the spatiotemporal dynamic mapping result for flight simulation path planning.

[0055] This application has the following advantages:

[0056] By acquiring raw multispectral imagery and raw synthetic aperture radar data, and filtering target data with cloud coverage below a set threshold, the interference of clouds on image information can be reduced, ensuring the effectiveness of subsequent data processing. By extracting terrain edge features from target multispectral imagery, dynamically adjusting the radar beam incidence angle, and increasing the point cloud density in complex terrain areas, non-uniform resolution point cloud data is generated. This allows the point cloud data to retain more details in complex terrain areas, adapting to the characteristic representation needs of different terrains. By constructing a dual-branch processing model and weightedly fusing the terrain features output from the two branches, a fused feature is obtained. This combines the advantages of multispectral imagery and point cloud data, improving the completeness and accuracy of terrain features. By dynamically associating vegetation index distribution with elevation information, a spatiotemporal dynamic mapping result is generated, establishing the relationship between terrain and vegetation / crop types, reflecting their changes over time. By generating a semantically labeled 3D flight simulation terrain model based on the spatiotemporal dynamic mapping result, a virtual environment that conforms to actual terrain characteristics can be provided for flight simulation path planning.

[0057] Furthermore, by analyzing terrain edge features to generate terrain edge feature lines, sub-regions are divided and the dominant slope aspect is determined. The complexity of the sub-regions is calculated by combining the density of terrain edge feature lines, slope aspect coefficient, and complexity correction factor. Based on this, the radar beam incident angle is adjusted, the reflection point position is scanned, and an initial point cloud is generated. The point cloud density is increased by increasing the tilt of the radar beam incident angle in the terrain-complex area, generating point cloud data with non-uniform resolution. The above method can accurately quantify the terrain complexity, making the adjustment of the radar beam incident angle more in line with the terrain features, thereby allowing the point cloud data to form a differentiated distribution in terrain regions with different levels of complexity. While ensuring data processing efficiency, it improves the detail richness of the point cloud data in terrain-complex areas.

[0058] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 A flowchart illustrating a multispectral fusion method for generating flight simulation terrain, provided in an embodiment of this application;

[0061] Figure 2 A schematic diagram of the structure of a multispectral fusion flight simulation terrain generation system provided in this application embodiment;

[0062] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0063] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0064] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] To address the low accuracy of flight simulation terrain generation in existing technologies due to low-quality image data, difficulty in accurately reflecting the true undulations of terrain, and lack of vegetation information, this application provides a multispectral fusion method for generating flight simulation terrain. This method employs the following design concept: First, multi-band images and radar detection data of the terrain to be detected are collected, selecting images with less cloud cover and corresponding radar data. The edge shapes of the terrain are identified from the images, and the radar detection angle is adjusted based on these shapes to make the 3D point data denser in complex terrain areas. The multi-band images and 3D point data are processed separately through two branches, and then the terrain features of both are combined. The growth index distribution of surface vegetation is correlated with the terrain elevation information to form a correlation result that reflects the changes in plant type and terrain over time. Finally, based on this result, a 3D terrain model with detailed labels is generated for flight simulation path planning. This approach reduces the impact of clouds, making the details of complex terrain areas clearer, and combines vegetation information to generate a more realistic flight simulation terrain model.

[0067] Figure 1 A flowchart of a multispectral fusion method for generating flight simulation terrain provided in this application embodiment is shown below. Figure 1 As shown, the method includes:

[0068] S11. Collect original multispectral images and original synthetic aperture radar data of the terrain area to be measured. Identify cloud coverage based on the pixel value distribution of the original multispectral images. Filter out target multispectral images and target synthetic aperture radar data of the corresponding time window with cloud coverage below a set threshold.

[0069] The original multispectral image can be a terrain image containing information on various wavelengths of light, reflecting the reflection of different light rays on the terrain surface; the original synthetic aperture radar (SAR) data is data obtained by radar equipment through terrain detection, reflecting the three-dimensional structure information of the terrain; cloud coverage is the proportion of the area occupied by clouds in the original multispectral image; the target multispectral image is the original multispectral image with a cloud coverage rate lower than a set proportion; the target SAR data is the original SAR data acquired around the same time as the target multispectral image; the result of processing these data is the filtered target multispectral image and the corresponding target SAR data.

[0070] In this embodiment, the original multispectral image and original synthetic aperture radar (SAR) data of the terrain area to be detected are first collected. Then, the distribution of pixel values ​​in the original multispectral image is observed. Since clouds usually appear as areas with high pixel values ​​in the image, the proportion of the total area of ​​these high pixel value areas to the total area of ​​the image is calculated to determine the cloud coverage rate. Finally, an upper limit for cloud coverage rate is set, and original multispectral images with coverage rates lower than the upper limit are selected as target multispectral images. SAR data with shooting times close to these target images are then found as target SAR data. For example, when processing data for a certain area, if the total area of ​​the original multispectral image is 1000 square meters, and the area of ​​the cloud region with high pixel values ​​is 200 square meters, then the cloud coverage rate is 200 divided by 1000, which equals 20%. If the upper limit is set to 30%, then this image can be used as the target multispectral image. SAR data with a shooting time difference of no more than 1 hour is then found as target SAR data.

[0071] S12. Extract terrain edge features from the target multispectral image, and dynamically adjust the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution. During the generation process, increase the point cloud density in complex terrain areas by increasing the tilt of the radar beam incident angle. Complex terrain areas include steep slopes and canyon areas.

[0072] Among them, target multispectral image is the selected target multispectral image; terrain edge features are the features at the junction of different regions in the terrain, such as the boundary between mountains and plains, the two sides of the canyon, etc.; radar beam incidence angle is the angle between the detection signal emitted by the radar equipment and the terrain surface; point cloud data is a data set composed of a large number of three-dimensional points, each point representing a location on the terrain surface; non-uniform resolution point cloud data is the case where the density of points in the point cloud varies in different terrain regions; complex terrain regions are regions with large terrain undulations and complex structures, such as steep slopes and canyons; the result generated after processing is non-uniform resolution point cloud data.

[0073] In this embodiment, terrain edge features are first extracted from the target multispectral image. Specifically, this involves observing locations where pixel values ​​suddenly change, such as the boundary between flat land with low pixel values ​​and mountainous terrain with high pixel values, to determine the terrain edge. Secondly, the radar beam's incident angle is adjusted based on the extracted terrain edge features. For areas with complex terrain, the tilt of the radar beam's incident angle is increased, allowing the radar signal to more comprehensively detect these areas. Then, the terrain is detected using the adjusted radar beam, and the reflected signals are collected and converted into point cloud data. Finally, because the radar beam's incident angle tilt is greater in areas with complex terrain, it will... Obtaining more reflected signals results in a greater number and density of points in the point cloud data of these areas, forming point cloud data with non-uniform resolution. For example, in the processing of a certain area, if the edge of a steep slope is found in the image, and the tilt angle of the radar beam incident angle in that area is adjusted from 30 degrees to 50 degrees, after detection, the number of points per square meter in the steep slope area is 70, while the surrounding flat land has 30 points per square meter. This is because the radar signal reflects more points after the tilt angle increases. The calculation method is to divide the total number of points in the area by the area of ​​the area. For example, if the area of ​​the steep slope area is 10 square meters and the total number of points is 700, 700 divided by 10 equals 70 points per square meter.

[0074] S13. Construct a dual-branch processing model. The first branch processes the target multispectral image, and the second branch processes the point cloud data. The high-resolution terrain features output by the two branches are fused through a weighted mechanism to obtain the fused features.

[0075] The dual-branch processing model comprises two processing paths, each handling different types of data. The first branch specifically processes the target multispectral imagery, while the second branch processes point cloud data. The weighting mechanism assigns different weights based on the reliability of different data before merging them. The fused feature is a composite feature obtained by merging the terrain features from the two branches. The final result is the fused feature. For example, the dual-branch processing model can be constructed based on a two-stream optimized U-Net model.

[0076] In this embodiment, a dual-branch processing model is first constructed. The first branch processes the target multispectral image, specifically analyzing pixel distribution, color changes, etc., to extract terrain-related features, such as texture features of vegetation-covered areas. The second branch processes point cloud data, analyzing the distribution and distance of three-dimensional points in the point cloud to extract terrain undulation features. Then, the weights of the target multispectral image and point cloud data are determined according to their respective reliability. For example, the image is more reliable for vegetation identification, so the weight is set to 0.6, and the point cloud is more reliable for terrain undulation, so the weight is set to 0.4. Finally, the features extracted by the first branch and the features extracted by the second branch are merged according to the determined weights to obtain a fused feature. For example, in the processing of a certain area, the texture feature value extracted by the first branch is 0.8, and the terrain undulation feature value extracted by the second branch is 0.6, with weights of 0.5 and 0.5 respectively. Then, the fused feature value is 0.8 multiplied by 0.5 plus 0.6 multiplied by 0.5 equals 0.7, thus combining the information of the two features.

[0077] S14. Dynamically associate the vegetation index distribution of the topographic region to be measured with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of the vegetation cover area topography and crop type.

[0078] Among them, the vegetation index distribution is a numerical distribution reflecting the vegetation growth in the terrain area. The higher the value, the more vigorous the vegetation growth. The elevation information is the height information of each location in the terrain area. The spatiotemporal dynamic mapping result is a result that reflects the correspondence between the terrain and crop type of the vegetation-covered area over time. The result generated after processing is the spatiotemporal dynamic mapping result.

[0079] In this embodiment, the vegetation index distribution of the terrain area is first calculated based on the pixel values ​​of different bands in the target multispectral image. For example, the pixel values ​​of two bands are compared, and the difference between the pixel value of band 1 and the pixel value of band 2 is divided by the sum of the pixel values ​​of band 1 and band 2 to obtain the vegetation index value at each location. Next, elevation information is extracted from the fused features to determine the height value at each location. Then, the vegetation index distribution and elevation information are correlated according to spatial location, so that the vegetation index value at each location matches the height value at that location. Finally, the correlation between the two is adjusted according to the changes in the height value. The correlation between vegetation index and altitude is stronger in areas with large changes in altitude and weaker in areas with small changes in altitude. Finally, the correlation results at different time points are recorded. Combined with the growth of different crops at different altitudes and vegetation indices, the correspondence between vegetation cover area topography and crop type over time is generated, resulting in a spatiotemporal dynamic mapping result. For example, at a certain location, the pixel value of band 1 is 80, the pixel value of band 2 is 20, and the vegetation index is 80 minus 20 divided by the sum of 80 plus 20, which equals 0.6. The elevation of this location is 200 meters. In spring, this correlation corresponds to crop E. In summer, the vegetation index becomes 0.7, but it still corresponds to crop E.

[0080] S15. Based on the spatiotemporal dynamic mapping results, generate a three-dimensional flight simulation terrain model with semantic labels for flight simulation path planning.

[0081] Among them, the semantically labeled 3D flight simulation terrain model is a virtual 3D terrain model containing category labels such as terrain, vegetation, and crop type; flight simulation path planning is to design a suitable flight route in the virtual terrain model; the processed result is a semantically labeled 3D flight simulation terrain model, which is used for flight simulation path planning.

[0082] In this embodiment, a three-dimensional terrain model is first constructed based on the spatiotemporal dynamic mapping results obtained earlier. This model can reflect the undulations of the terrain. Then, semantic labels are added to the three-dimensional model, such as labeling steep slope areas as steep slopes and planting areas of crop E as crop E. Then, this three-dimensional model with semantic labels is used as the flight simulation terrain model. Finally, this model is used for flight simulation path planning. Based on the labels such as steep slopes and high-altitude crop E (high altitude) in the model, a flight route is planned to avoid these areas. For example, in a certain model, canyons (complex terrain), crop E area (2 meters high), and flat land are labeled. When planning the route, the distance and safety factor of each area from the starting point to the end point are calculated, avoiding canyons and crop E areas with low safety factors, and selecting flat land with high safety factors.

[0083] For example, when generating terrain for region B using flight simulation, the original multispectral imagery and original synthetic aperture radar (SAR) data for that region are first acquired. The cloud coverage of the original multispectral imagery is calculated. One image has a total area of ​​500 square meters, with a cloud area of ​​100 square meters, resulting in a coverage rate of 100 ÷ 500 = 20%. A threshold of 30% is set, therefore this imagery is selected as the target multispectral imagery. Furthermore, SAR data captured within one hour of each other are selected as the target SAR data. Next, terrain edge features were extracted from the target multispectral image, revealing a 20-square-meter canyon and a 15-square-meter steep slope. Based on these edge features, the radar beam incident angle tilt for these two areas was adjusted from 30 degrees to 50 degrees. After detection, the canyon had a total of 1600 points (1600 ÷ 20 = 80 points / square meter), the steep slope had a total of 1050 points (1050 ÷ 15 = 70 points / square meter), and the flat area of ​​100 square meters had a total of 3000 points (3000 ÷ 100 = 30 points / square meter), generating point cloud data with non-uniform resolution. Then, a two-branch model was used for processing. The first branch extracted texture features from the target multispectral image (value 0.7), and the second branch extracted terrain undulation features from the point cloud data (value 0.8), each assigned a weight of 0.5. The fused feature value was 0.7 × 0.5 + 0.8 × 0.5 = 0.75. Next, the vegetation index was calculated. In region C, the pixel value of band 1 was 100 and the pixel value of band 2 was 20, with an index of (100-20)÷(100+20)≈0.67, corresponding to an elevation of 200 meters. In region D, the pixel value of band 1 was 40 and the pixel value of band 2 was 30, with an index of (40-30)÷(40+30)≈0.14, corresponding to an elevation of 300 meters. Combining the time change records, in spring, region C corresponds to crop E and region D corresponds to crop F. Even after the index changes in summer, the same crop is still associated with the same crop, forming a spatiotemporal dynamic mapping result. Finally, a 3D model was constructed based on this result, labeled with tags such as "canyon," "crop area E (height 2 meters)," and "flat land." When planning the flight route, the canyon and crop area E were avoided, and the flat land area was selected to complete the entire process.

[0084] By executing steps S11-S15, this embodiment of the application processes the entire process, selecting high-quality multispectral imagery and radar data during the data acquisition stage, reducing cloud interference on terrain information, and ensuring data consistency over time. During the point cloud data generation stage, the radar detection angle is dynamically adjusted to make the point clouds in complex terrain areas denser, preserving details of key areas while avoiding data redundancy. The fusion processing of the dual-branch model integrates the texture information of the multispectral imagery and the three-dimensional structural information of the point cloud data, making terrain features more comprehensive and accurate. The dynamic correlation between vegetation index and elevation reflects the spatiotemporal changes in terrain, vegetation, and crop types, enriching the semantic information of the model. The final generated three-dimensional model with semantic labels provides a reliable reference for flight simulation path planning, balancing terrain realism and flight safety, and overall improving the accuracy and practicality of flight simulation terrain generation.

[0085] In one possible embodiment, point cloud data with uniform resolution is prone to data redundancy in simple terrain areas and lack of detail in complex areas. Generating point cloud data with non-uniform resolution can make the points in complex areas denser to retain details and the points in simple areas fewer to reduce redundancy, balancing accuracy and efficiency. Therefore, in S12, the radar beam incidence angle is dynamically adjusted based on terrain edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density in complex terrain areas is increased by increasing the tilt of the radar beam incidence angle, including:

[0086] Step 121: By analyzing the terrain edge features, generate the outline of the terrain area to be measured, calculate the curvature change rate of the outline, and generate terrain edge feature lines based on the curvature change rate.

[0087] Among them, the topographic edge features are the features at the junction of different areas in the topography (such as flat land and mountains, and the sides of canyons); the outline of the topographic area to be measured is the approximate outer boundary of the topography in that area; the rate of curvature change is the speed at which the curvature of each point on the outline changes; the topographic edge feature line is a line generated based on the rate of curvature change that can reflect the specific location and direction of the topographic edge, and the result generated after processing is the topographic edge feature line.

[0088] In this embodiment, the terrain edge features are first analyzed. By identifying the location and shape of the boundary between different areas, the outline of the terrain area to be measured is delineated. Then, the rate of curvature change of each point on the outline is calculated. Specifically, the change in tangent direction of adjacent points on the outline is measured to determine the change in the degree of curvature at each point. Finally, based on the magnitude of the rate of curvature change, the positions with significant changes are connected to generate terrain edge feature lines. For example, in area A, the boundary features between flat land and steep slope are first analyzed to delineate the outline. The tangent directions of two adjacent points on a certain segment of the outline are calculated to be 30 degrees and 60 degrees respectively, with an angle difference of 30 degrees and a distance of 2 meters between the two points. The rate of curvature change is 30 degrees ÷ 2 meters = 15 degrees / meter. The positions with large rates of curvature change are connected to generate the terrain edge feature lines of the steep slope.

[0089] Step 122: Based on the spatial distribution pattern of terrain edge feature lines, divide the terrain area to be measured into multiple sub-regions with similar terrain features, and determine the corresponding dominant slope aspect of each sub-region.

[0090] Among them, the spatial distribution pattern of the terrain edge feature lines is the distribution density and direction of the feature lines in the terrain area to be measured; the sub-region is a small area with similar terrain features that is divided from the terrain area to be measured; the dominant slope aspect is the main tilt direction of the terrain in each sub-region. The result generated after processing is the divided sub-regions and the dominant slope aspect of each sub-region.

[0091] In this embodiment, the spatial distribution patterns (such as density and orientation) of terrain edge feature lines are first observed; then, based on these patterns, the terrain area to be measured is divided into multiple sub-regions, ensuring that the distribution and orientation of feature lines are similar within each sub-region; finally, the terrain slope direction of each sub-region is analyzed to determine the dominant slope aspect. For example, in region A, it is found that the feature lines are dense in the northeast and run in a northeast-southwest direction, while they are sparse in the southwest and run in an east-west direction. Based on this, the region is divided into northeastern and southwestern sub-regions. The terrain in the northeast mainly slopes to the southeast, and the dominant slope aspect is southeast.

[0092] Step 123: Based on the density value of the terrain edge feature line and the corresponding dominant slope aspect of each sub-region, calculate the complexity value of each sub-region, and adjust the radar beam incident angle according to the complexity value.

[0093] Among them, the density value of terrain edge feature lines is the total length of terrain edge feature lines within a unit area; the complexity value is a value that quantifies the complexity of the terrain in a sub-region; the radar beam incident angle is the angle between the radar transmitted signal and the terrain surface, and the processed result is to adjust the radar beam incident angle according to the complexity value.

[0094] In this application example, the topographic edge feature line density value (total length ÷ sub-region area) of each sub-region is first calculated; then, a correction coefficient is determined based on the dominant slope aspect (the coefficient varies depending on the angle between the dominant slope aspect and a specific direction); next, the density value is multiplied by the correction coefficient to obtain the complexity value; finally, the radar beam incident angle is adjusted according to the value, with a larger value indicating a greater inclination. For example, in the central sub-region of region A, the total length of the feature lines is 100 meters, the area is 50 square meters, the density value is 100 ÷ 50 = 2 meters / square meter, the correction coefficient is 1.2, and the complexity value is 2 × 1.2 = 2.4, thus adjusting the inclination from 30 degrees to 45 degrees.

[0095] Step 124: Control the radar to scan each sub-region according to the adjusted radar beam incident angle, obtain the reflection signal of each sub-region, analyze the return time and strength of the reflection signal to determine the position of the first reflection point.

[0096] Among them, radar is a device that transmits and receives detection signals; reflected signal is the radar signal reflected after encountering terrain; return time is the time from transmission to reception; strength is the intensity of the reflected signal; and the location of the first reflection point is the spatial location of the terrain surface point from which the reflected signal originally originated, which is obtained after processing.

[0097] In this embodiment, the radar is first controlled to scan a sub-region at an adjusted incident angle and transmit a detection signal; then, reflected signals are received; next, the return time (longer indicates greater distance) and intensity (reflecting terrain reflectivity) of the signals are analyzed; finally, the position of the first reflection point is calculated based on the return time and intensity. For example, in the central sub-region of region A, the radar transmits a signal at a 45-degree tilt angle, and a certain reflected signal has a return time of 0.002 seconds and a signal velocity... meters per second, distance is ×0.002÷2=300 meters, and the location is determined as (200,300,150) based on the strength.

[0098] Step 125: Generate an initial point cloud data set based on the position of the first reflection point. In the initial point cloud data, for areas with complex terrain, increase the point cloud density in the areas with complex terrain by increasing the tilt of the radar beam incident angle, so as to generate point cloud data with non-uniform resolution.

[0099] The initial point cloud data set is the set of all first reflection point locations; the complex terrain area is an area with large terrain undulations and complex structure (such as steep slopes and canyons); the radar beam incident angle tilt is the tilt angle between the radar beam and the terrain surface; the non-uniform resolution point cloud data is point cloud data with different point densities in different terrain areas, and the result generated after processing is non-uniform resolution point cloud data.

[0100] In this embodiment, all first reflection point locations are first integrated to form an initial point cloud data set; then, complex terrain areas are identified; these areas are then scanned again with the radar incident angle increased to obtain more reflection points; finally, the newly added points are added to the initial set to increase the point density in complex areas, forming a non-uniform point cloud. For example, the initial point density in the middle of region A (complex area) is 50 points / square meter, and after increasing the incident angle, an additional 30 points / square meter are added, reaching a density of 80 points / square meter, while the density in simple areas remains at 30 points / square meter.

[0101] For example, in the terrain processing of area A, the terrain edge features (such as the boundary between canyons and mountains) are analyzed first to outline the region. The curvature change rate of each point on the outline is calculated (60 degrees / meter at the edge of the canyon, 0.2 degrees / meter on flat land). Points with large curvature change rates are connected to generate terrain edge feature lines. Then, the distribution of feature lines is observed (dense in the middle and running north-south, sparse in the east-west and running east-west). Area A is divided into central, eastern and western sub-regions. The dominant slope direction in the central region is determined to be east-west, in the eastern region to be south, and in the western region to be north. Next, calculate the characteristic line density of each sub-region: the central region has a total length of 240 meters and an area of ​​80 square meters, with a density of 240 ÷ 80 = 3 meters / square meter. Combining this with a dominant slope aspect correction factor of 1.1 and a complexity value of 3 × 1.1 = 3.3, adjust the radar incident angle to 50 degrees. The eastern region has a density of 60 ÷ 40 = 1.5 meters / square meter, a correction factor of 1.0, and a value of 1.5, so adjust the incident angle to 35 degrees. Then, control the radar to scan at the adjusted angle and receive reflected signals. A certain signal returned in 0.003 seconds, at a distance of 3 × 10... 8 ×0.003÷2=450 meters, determine the location of the first reflection point (150,250,180); finally, integrate the reflection points to form an initial point cloud, with 4800 points (60 points / square meter) in the central 80 square meters and 2500 points (25 points / square meter) in the eastern 100 square meters. Increase the incident angle of the central part to 60 degrees and scan again, adding 300 points, with a density of 60+300÷80=63.75 points / square meter. Keep the original density in the eastern part to generate non-uniform point cloud data.

[0102] By executing steps 121 to 125, this embodiment of the application generates terrain edge feature lines to accurately capture the location and direction of terrain edges, providing a basis for region division. The divided sub-regions and determined dominant slope aspects make terrain analysis more targeted. The computational complexity numerically quantifies the terrain complexity, and the radar incident angle adjusted accordingly makes the detection more closely match the terrain features. The obtained first reflection point location provides basic data for point cloud generation. The final generated non-uniform resolution point cloud data retains more details in complex terrain areas to reflect the real structure, and reduces data redundancy in simple areas to improve efficiency. Overall, it balances data accuracy and processing efficiency, providing high-quality basic data for subsequent flight simulation terrain modeling and path planning.

[0103] In one possible embodiment, step 123, calculating the complexity value of each sub-region based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region, includes:

[0104] Step a1: Calculate the total length of all terrain edge feature lines in each sub-region, divide the total length by the area of ​​the corresponding sub-region, and obtain the density value of terrain edge feature lines in the unit sub-region.

[0105] The total length of the terrain edge feature lines refers to the sum of the lengths of all terrain edge feature lines within each sub-region; the area of ​​the sub-region is the area occupied by each sub-region; the density value of the terrain edge feature lines within a unit sub-region is the value obtained by dividing the total length by the area of ​​the sub-region, which is used to reflect the density of the terrain edge feature lines within the sub-region. The result obtained after processing is this density value.

[0106] In this embodiment, the lengths of all terrain edge feature lines in each sub-region are first counted, and these lengths are added together to obtain the total length. Then, the area of ​​the sub-region is measured. Finally, the total length is divided by the area of ​​the corresponding sub-region to obtain the density value of the terrain edge feature lines in the unit sub-region. For example, in sub-region B, the lengths of all terrain edge feature lines are 50 meters, 80 meters, and 120 meters, respectively, and the total length is 50 + 80 + 120 = 250 meters. The area of ​​this sub-region is 200 square meters, and the density value is 250 ÷ 200 = 1.25 meters / square meter.

[0107] Step a2: Analyze the range of the angle between the dominant slope direction and the due north direction in each sub-region, and divide the range of the angle into multiple angle intervals.

[0108] Among them, the dominant slope direction is the main tilt direction of the terrain in the sub-region; due north is geographical due north; the included angle range is the angle range formed between the dominant slope direction and due north; the included angle interval is a series of continuous angle segments divided from the included angle range of 0-360 degrees, and the processed result is the divided included angle interval.

[0109] In this embodiment of the application, the dominant slope direction of each sub-region is first determined, and then the angle between the dominant slope direction and the due north direction is measured to obtain the included angle range. Finally, the included angle range of 0-360 degrees is divided into multiple included angle intervals according to fixed angle intervals (such as every 45 degrees). For example, the included angle between the dominant slope direction of sub-region C and the due north direction is 70 degrees. The 0-360 degrees is divided into intervals such as 0-45 degrees, 46-90 degrees, and 91-135 degrees. 70 degrees belongs to the included angle interval of 46-90 degrees.

[0110] Step a3: Determine the corresponding slope aspect coefficient based on the included angle range of the dominant slope aspect of each sub-region.

[0111] Among them, the included angle interval where the dominant slope aspect is located refers to the angle segment corresponding to the included angle between the dominant slope aspect of each sub-region and the due north direction; the slope aspect coefficient is a value set according to the included angle interval to correct the complexity of the terrain. Different intervals correspond to different coefficients, and the slope aspect coefficient of each sub-region is obtained after processing.

[0112] In this embodiment of the application, the dominant slope aspect of each sub-region is first determined to be within which angle range. Then, according to the preset correspondence (e.g., 1.0 for the 0-45 degree range and 1.2 for the 46-90 degree range), the slope aspect coefficient corresponding to the sub-region is determined. For example, the dominant slope aspect angle of sub-region C is within the 46-90 degree range, and the slope aspect coefficient corresponding to this range is 1.2. Therefore, the slope aspect coefficient of sub-region C is 1.2.

[0113] Step a4: Calculate the average curvature of the terrain edge feature lines in each sub-region, and divide the average curvature into multiple levels. Each level of the average curvature corresponds to a different complexity correction factor.

[0114] Among them, the average curvature of the terrain edge feature lines refers to the average curvature of all terrain edge feature lines in the sub-region. The curvature can be represented by the ratio of the actual length of the line to the straight distance. The level is a series of continuous intervals that divide the average curvature. The complexity correction factor is a value set according to the level to correct the complexity of the terrain. Different levels correspond to different factors. After processing, the complexity correction factor of each sub-region is obtained.

[0115] In this embodiment, the curvature of each terrain edge feature line in each sub-region is first calculated (actual length ÷ straight distance). Then, these curvatures are added together and divided by the number of feature lines to obtain the average curvature. Next, the average curvature is divided into multiple levels (e.g., 1.0-1.5 is level 1, 1.6-2.0 is level 2). Finally, the complexity correction factor is determined according to the preset value corresponding to the level (e.g., level 1 corresponds to 1.0, level 2 corresponds to 1.3). For example, the curvatures of the two feature lines in sub-region D are 1.4 and 1.6 respectively, and the average curvature is (1.4+1.6)÷2=1.5, which belongs to level 1.0-1.5, and the corresponding complexity correction factor is 1.0.

[0116] Step a5: The product of the density value of the terrain edge feature line in the unit sub-region, the corresponding slope aspect coefficient, and the corresponding complexity correction factor is used as the complexity value of each sub-region.

[0117] Among them, the density value of the terrain edge feature lines in the unit sub-region is a value that reflects the density of the edge lines; the aspect coefficient is a value that corrects the influence of the aspect; the complexity correction factor is a value that corrects the influence of the line curvature; the complexity value is the product of these three values, which is used to comprehensively reflect the terrain complexity of the sub-region. The result after processing is the complexity value of each sub-region.

[0118] In this embodiment of the application, the density value obtained in step a1, the aspect coefficient obtained in step a3, and the complexity correction factor obtained in step a4 are first obtained. Then, these three values ​​are multiplied together, and the result is the complexity value of the corresponding sub-region. For example, the density value of sub-region D is 1.8 m / m², the aspect coefficient is 1.1, and the complexity correction factor is 1.0. Then the complexity value is 1.8 × 1.1 × 1.0 = 1.98.

[0119] For example, when processing sub-region E of region A, in step a1, the lengths of the three terrain edge feature lines within this sub-region are calculated to be 60 meters, 90 meters, and 150 meters, respectively, with a total length of 60 + 90 + 150 = 300 meters. The area of ​​the sub-region is 200 square meters, and the density value is 300 ÷ 200 = 1.5 meters / square meter. In step a2, the angle between the dominant slope aspect and due north is determined to be 60 degrees, and the angle range is divided into intervals such as 0-45 degrees and 46-90 degrees. 60 degrees belongs to the 46-90 degree interval. In step a3, the slope aspect coefficient corresponding to this interval is 1.2. In step a4, the curvature of the three feature lines is calculated to be 1.6, 1.8, and 2.0, respectively, with an average curvature of (1.6 + 1.8 + 2.0) ÷ 3 = 1.8. The classification level is 1.6-2.0, corresponding to level 2, with a correction factor of 1.3. In step a5, the complexity value is 1.5 × 1.2 × 1.3 = 2.34.

[0120] By executing steps a1 to a5, this embodiment of the application calculates density values ​​step by step, divides slope aspect intervals, determines slope aspect coefficients, calculates average curvature and complexity correction factors, and finally multiplies the three to obtain a complexity value. It comprehensively considers the density of terrain edges, slope aspect characteristics and line curvature characteristics, making the quantification result of complexity more in line with actual terrain characteristics, accurately reflecting the terrain complexity of sub-regions, providing a scientific and reliable basis for subsequent adjustment of relevant parameters according to complexity, and improving the accuracy and pertinence of terrain analysis.

[0121] In one possible instance, step 125 involves generating an initial point cloud data set based on the location of the first reflection point. Within this initial point cloud data, for areas with complex terrain, the point cloud density is increased by increasing the tilt of the radar beam incident angle to generate point cloud data with non-uniform resolution, including:

[0122] b1. Arrange the position information of all first reflection points according to three-dimensional spatial coordinates to generate an initial point cloud data set.

[0123] Among them, the location information of the first reflection point is the three-dimensional spatial coordinates of the point corresponding to the signal reflected from the terrain surface detected by the radar for the first time. The three-dimensional spatial coordinates are numerical values ​​that describe the position of the point in three-dimensional space. The initial point cloud data set is a dataset formed by arranging the location information of all the first reflection points according to the three-dimensional spatial coordinates. It is used to initially reflect the three-dimensional structure of the terrain. The result after processing is the initial point cloud data set.

[0124] In this embodiment, the location information of all first reflection points is first collected. This information includes the specific values ​​of each point in the x, y, and z directions. Then, these points are arranged in the order of the size of the three-dimensional spatial coordinates (first by the x coordinate, then by the y coordinate, and finally by the z coordinate). Finally, the arranged points are integrated to form an initial point cloud data set. For example, in region A, the coordinates of three first reflection points are obtained as (10, 20, 5), (15, 25, 8), and (8, 18, 6). After arranging them in ascending order of the x coordinate, an initial point cloud data set containing these three points is generated.

[0125] b2. Assign each initial point cloud data in the initial point cloud dataset to the corresponding sub-region.

[0126] The initial point cloud data set is a dataset composed of the three-dimensional coordinates of the first reflection point. The initial point cloud data is the three-dimensional coordinates of a single point in the set. The sub-region is a pre-divided small region with similar terrain features. The corresponding sub-region refers to the small region where the initial point cloud data is located (i.e., the coordinates of the point fall within the range of the sub-region). The result after processing is the initial point cloud data allocated to each sub-region.

[0127] In this embodiment, the range of each sub-region (i.e., the start and end values ​​of the x and y coordinates) is first determined. Then, the x and y coordinates of each point in the initial point cloud data set are examined one by one to determine which sub-region it falls within. Finally, the point is assigned to the corresponding sub-region. For example, the x range of sub-region B in region A is 0-200 and the y range is 0-200. The coordinates of a point are (150, 180, 10). Its x and y coordinates fall within sub-region B, so it is assigned to sub-region B.

[0128] b3. Extract the sub-regions that are considered to be complex terrain regions from all sub-regions that have been allocated initial point cloud data. Control the target synthetic aperture radar to scan the complex terrain regions multiple times according to the target radar beam incident angle tilt to obtain the location information of the second reflection point. The target radar beam incident angle tilt is greater than the adjusted radar beam incident angle.

[0129] The initial point cloud data consists of the coordinates of the first reflection point assigned to the sub-region. The sub-region with complex terrain is one with large terrain undulations and complex structure. The target synthetic aperture radar is a device used to transmit and receive detection signals. The target radar beam incident angle tilt is a tilt angle that is larger than the previously adjusted angle (used to increase the point cloud density). The location information of the second reflection point is the three-dimensional coordinates of the reflection point newly acquired after multiple scans. The result obtained after processing is the location information of the second reflection point.

[0130] In this embodiment, firstly, from the sub-regions allocated with initial point cloud data, sub-regions with complex terrain are identified based on the complexity value. Then, the target synthetic aperture radar is controlled to scan these regions multiple times using a target radar beam incident angle tilt that is larger than before. Each scan acquires new reflection signals. Finally, the location information of the second reflection points is determined based on these signals. For example, sub-region B of region A is a region with complex terrain. The previous tilt was 50 degrees, and the target tilt was set to 65 degrees. After scanning it twice, the coordinates of 25 second reflection points are obtained.

[0131] b4. Add the location information of the second reflection point to the initial point cloud data of the corresponding complex terrain area to obtain the target point cloud data of the complex terrain area, so as to improve the point cloud density of the complex terrain area.

[0132] The location information of the second reflection point is the newly acquired three-dimensional coordinates of the reflection point. The initial point cloud data of the complex terrain area is the coordinates of the original first reflection point in the area. The target point cloud data is the set of coordinates of all reflection points in the area after the addition of the second reflection point. The point cloud density is the number of reflection points per unit area. The result after processing is the target point cloud data of the complex terrain area.

[0133] In this embodiment of the application, the location information of the second reflection point obtained in step b3 is first collected, and then these coordinates are added to the initial point cloud data of the corresponding complex terrain area and merged with the original first reflection point coordinates. Finally, the target point cloud data of the area is obtained. Due to the addition of reflection points, the point cloud density (total number of points ÷ area of ​​the region) is improved. For example, the B sub-region of region A initially has 10 points (area of ​​20 square meters, density of 0.5 points / square meter). After adding 25 second reflection points, the total number of points is 35, and the density becomes 35 ÷ 20 = 1.75 points / square meter.

[0134] b5. By comparing the point cloud density of each sub-region, it is verified that the point cloud density of the terrain complex region is higher than that of the non-terrain complex region in the terrain region to be tested. After the verification is passed, point cloud data with non-uniform resolution is generated based on the target point cloud data of the terrain complex region and the initial point cloud data of the non-terrain complex region.

[0135] Point cloud density is the number of reflection points per unit area. Point cloud density in terrain-complex areas is the density of target point cloud data in that area. Non-terrain-complex areas are sub-regions with gentle terrain. Their initial point cloud data is the coordinates of the first reflection point. Non-uniform resolution point cloud data is the dataset where the point cloud density in terrain-complex areas is higher than that in non-terrain-complex areas. The result after processing is non-uniform resolution point cloud data.

[0136] In this embodiment, the point cloud density of each sub-region is first calculated (target point cloud data is used for complex terrain regions, and initial point cloud data is used for non-complex regions). Then, these densities are compared to verify whether the density of complex terrain regions is higher than that of non-complex regions. After verification, the target point cloud data of complex terrain regions is integrated with the initial point cloud data of non-complex regions to generate point cloud data with non-uniform resolution. For example, the density of sub-region B (complex) in region A is 1.75 points / square meter, and the density of sub-region C (non-complex) is 0.25 points / square meter. After comparison, the density of B is higher. After integration, a non-uniform dataset is formed.

[0137] For example, in the terrain processing of area A, step b1 first collects the coordinates of the first reflection point (10,20,5), (15,25,8), etc., and arranges them according to three-dimensional coordinates to form an initial point cloud data set; step b2 assigns these points to sub-regions, with 10 points in sub-region B (area 20 square meters) and 5 points in sub-region C (area 20 square meters); step b3 identifies sub-region B as a complex terrain area, scans it twice with a 65-degree tilt (greater than the previous 50 degrees), and obtains 25 second reflection points; step b4 adds these 25 points to sub-region B, with a total of 35 points and a density of 35÷20=1.75 points / square meter; step b5 calculates the density of sub-region C as 5÷20=0.25 points / square meter, and after comparison, the density of B is higher (verification passed), and integrates the data of the two regions to generate point cloud data with non-uniform resolution.

[0138] By executing steps b1 to b5, this embodiment of the application forms an initial point cloud by arranging the coordinates of reflection points and assigns it to sub-regions, providing a foundation for targeted processing; for complex terrain areas, multiple scans with a greater tilt are used to obtain more reflection points, directly increasing the point cloud density; through verification, it is ensured that the density of complex areas is higher than that of simple areas. The resulting non-uniform resolution point cloud data can not only meticulously reflect the details of complex terrain, but also avoid data redundancy in simple areas, balancing data accuracy and processing efficiency, and providing high-quality data support for subsequent 3D terrain modeling.

[0139] In one possible embodiment, the high-resolution terrain features include texture information and geometric information. S13: Construct a dual-branch processing model. The first branch processes the target multispectral image, and the second branch processes the point cloud data. The high-resolution terrain features output from the two branches are fused using a weighted mechanism to obtain fused features, including:

[0140] Step 131: Based on the first branch, decompose the target multispectral image into single-band images corresponding to multiple bands, determine the spatial distribution pattern of pixel values ​​in each single-band image, and extract texture information corresponding to terrain features based on the spatial distribution pattern.

[0141] The first branch is the part that processes the target multispectral image. The target multispectral image is a terrain image containing multiple wavelengths of light. The single-band image is the image corresponding to a single wavelength after the multispectral image is decomposed. The spatial distribution pattern of pixel values ​​is the arrangement and variation law of pixel values ​​in the single-band image. The texture information is the surface features related to the terrain features extracted based on this pattern. The result after processing is the texture information corresponding to the terrain features.

[0142] In this embodiment, the target multispectral image is first decomposed into multiple single-band images according to different light wavelengths based on the first branch. Then, the distribution and changes of pixel values ​​in each single-band image are observed to determine its spatial distribution pattern. Finally, based on the aggregation or dispersion characteristics of pixels in the pattern, texture information corresponding to the terrain features is extracted. For example, the multispectral image of region A is decomposed into red and blue single-band images. In the red band, the pixel values ​​of flat land are evenly distributed, while the pixel values ​​of steep slopes vary greatly. Based on this, texture information reflecting the roughness of different terrain surfaces is extracted.

[0143] Step 132: Based on the second branch, organize the point cloud data into multiple data groups according to spatial location relationships, analyze the distribution density and relative location relationships of the point cloud data in each data group, and extract the geometric information corresponding to the terrain features.

[0144] The second branch is the part that processes point cloud data. Point cloud data is terrain data composed of a large number of three-dimensional point coordinates. Data sets are subsets of point clouds organized according to spatial location (such as proximity). Distribution density is the number of points per unit volume within a data set. Relative positional relationship is the distance and orientation relationship between points within the set. Geometric information is the feature related to the three-dimensional structure of the terrain extracted based on density and position. The result after processing is the geometric information corresponding to the terrain features.

[0145] In this embodiment, the point cloud data is first grouped according to spatial location relationships based on the second branch (e.g., points that are close to each other are grouped together) to form multiple data groups. Then, the point distribution density (total number of points divided by the spatial volume within the group) and the relative positions between points (e.g., height difference, distance) of each data group are analyzed. Finally, geometric information corresponding to the terrain features is extracted based on this information. For example, in the point cloud data of region B, a certain data group has dense points and a height difference of 30 meters, and geometric information reflecting the steep terrain of the region is extracted.

[0146] Step 133: Determine the confidence weight of texture information based on the shooting parameters of the target multispectral image, and determine the confidence weight of geometric information based on the acquisition parameters of point cloud data.

[0147] Among them, the shooting parameters are the conditions when shooting multispectral images (such as illumination and device resolution), the acquisition parameters are the conditions when acquiring point cloud data (such as radar accuracy and number of scans), the confidence weight is a value that measures the reliability of information, the confidence weight of texture information is a weight determined based on the shooting parameters, the confidence weight of geometric information is a weight determined based on the acquisition parameters, and the result after processing is the confidence weight of texture information and geometric information respectively.

[0148] In this embodiment, the reliability weight of texture information is first determined based on the shooting parameters of the target multispectral image. If the lighting is sufficient and the device resolution is high, the texture information is more reliable and the weight is high; otherwise, the weight is low. Then, the reliability weight of geometric information is determined based on the acquisition parameters of point cloud data. If the radar accuracy is high and the number of scans is high, the geometric information is more reliable and the weight is high; otherwise, the weight is low. For example, when the multispectral image of region A is captured, the lighting is good, so the weight of texture information is set to 0.5. When the point cloud data is acquired, the radar accuracy is high, so the weight of geometric information is set to 0.5.

[0149] Step 134: Weight the texture information and geometric information according to their corresponding confidence weights to obtain the fused features.

[0150] Among them, texture information is the terrain surface features extracted from multispectral imagery, geometric information is the three-dimensional structural features extracted from point cloud data, confidence weight is a value that measures the reliability of the two, weighted merging is the process of combining the two types of information according to their weights, fusion feature is the comprehensive terrain feature obtained after merging, and the result after processing is the fusion feature.

[0151] In this embodiment, texture information and geometric information are first obtained, then the texture information is multiplied by its confidence weight, the geometric information is multiplied by its confidence weight, and finally the two results are added together to obtain the fused feature. For example, the texture information quantization value is 0.8 (weight 0.5), the geometric information quantization value is 0.7 (weight 0.5), and the fused feature is 0.8×0.5+0.7×0.5=0.75, which comprehensively reflects the surface and structural features of the terrain.

[0152] For example, when processing terrain data in area A, step 131 decomposes the multispectral image into red and green single-band images. In the red band, the pixels on flat land are evenly distributed, while the pixels on steep slopes vary greatly, and texture information is extracted (quantization value 0.8). Step 132 groups the point cloud data by location. A certain group of points is dense and has a large height difference, and geometric information is extracted (quantization value 0.7). Step 133 determines that the confidence weights of texture and geometric information are both 0.5 based on the shooting parameters (good lighting) and acquisition parameters (high radar accuracy). Step 134 calculates the fusion feature as 0.8×0.5+0.7×0.5=0.75, and obtains the fusion feature that integrates the two types of information.

[0153] By executing steps 131-134, this embodiment of the application extracts texture information of the terrain surface through the first branch and geometric information of the terrain structure through the second branch. It assigns reasonable confidence weights to the shooting and acquisition parameters and finally obtains the fused features through weighted merging. This not only preserves the detailed features of the terrain surface but also integrates the three-dimensional structural information, making the description of terrain features more comprehensive. At the same time, the introduction of confidence weights allows information with high reliability to account for a larger proportion in the fusion, improving the accuracy of the fused features and providing high-quality basic data for the subsequent generation of accurate three-dimensional terrain models.

[0154] In one possible embodiment, step 131 involves decomposing the target multispectral image into multiple single-band images corresponding to each band based on the first branch, determining the spatial distribution pattern of pixel values ​​in each single-band image, and extracting texture information corresponding to terrain features based on the spatial distribution pattern, including:

[0155] c1. Divide the target multispectral image into multiple independent single-band images according to the preset light wavelength range.

[0156] Among them, the target multispectral image is a terrain image containing information on multiple light wavelengths, the preset light wavelength range is a pre-defined interval of different light wavelengths, and the single-band image is the image corresponding to a single interval after the multispectral image is divided according to the preset wavelength range. The result generated after processing is multiple independent single-band images.

[0157] In this embodiment of the application, the target multispectral image is first divided into multiple independent single-band images according to a preset wavelength range. Each single-band image corresponds to a wavelength range. For example, the multispectral image of region A is divided into preset wavelength ranges of red light (600-700 nm) and green light (500-600 nm) to obtain a red single-band image and a green single-band image.

[0158] c2. Based on the values ​​of adjacent pixels in each single-band image, determine the positions where the value difference between adjacent pixels in each single-band image is greater than a preset difference threshold. According to the positions where the value difference is greater than the preset difference threshold, divide the corresponding single-band image and record the distribution characteristics of each region in the single-band image.

[0159] Among them, the numerical difference between adjacent pixels is the difference between the values ​​of two adjacent pixels in a single-band image, the preset difference threshold is a pre-set value to judge whether the pixel change is obvious, the position where the numerical difference is greater than the preset difference threshold is the boundary position where the pixel change is obvious, the distribution characteristics are the arrangement rules of pixel values ​​in each region after division, and the result generated after processing is the divided region and the distribution characteristics of each region.

[0160] In this embodiment, the numerical difference between adjacent pixels in each single-band image is first calculated. The difference is compared with a preset difference threshold to determine the positions where the difference is greater than the threshold. Based on these positions, the single-band image is divided into multiple regions, and the distribution characteristics of each region are recorded. For example, in the red single-band image of region A, the value of pixel A is 80, the value of adjacent pixel B is 50, the difference is 30, the preset difference threshold is 20, and 30 is greater than 20, so this position is determined as the boundary. Based on this, three regions are divided: the pixel value in region 1 is concentrated between 70 and 80, and the values ​​in region 2 are scattered.

[0161] c3. Based on the distribution characteristics of each region, determine the spatial distribution pattern of pixel values ​​in each single-band image.

[0162] Among them, the distribution characteristics of the region are the arrangement rules of pixel values ​​within the region (such as concentrated or dispersed), and the spatial distribution pattern of pixel values ​​is the overall pixel arrangement pattern of the region summarized based on the distribution characteristics (such as uniform or banded). The result generated after processing is the spatial distribution pattern of pixel values ​​in each single-band image.

[0163] In this embodiment, the distribution characteristics of each region in each single-band image are first sorted out. Based on these characteristics, the overall arrangement pattern of each region is summarized, and regions with similar patterns are merged to determine the spatial distribution pattern of pixel values ​​of the entire single-band image. For example, in the green light single-band image of region A, the pixel values ​​of region 1 are uniform and have small changes, while the values ​​of region 2 have large changes and are scattered. This is summarized as a spatial distribution pattern of "uniform distribution + scattered distribution".

[0164] c4. Based on the shape, topographic relief, and topographic morphology of each region in the spatial distribution pattern, extract the texture information corresponding to the topographic features from the single-band image.

[0165] Among them, the spatial distribution pattern is the overall arrangement rule of pixel values, the shape of the region is the shape of the region (such as block or strip), the terrain undulation is the change in the height of the terrain, the terrain morphology is the overall appearance of the terrain (such as flat or steep), and the texture information is the surface texture corresponding to the terrain features extracted from the distribution pattern. The result generated after processing is the texture information corresponding to the terrain features.

[0166] In this embodiment of the application, the shape, topographic relief and topographic morphology of each region in the spatial distribution pattern are first analyzed. Based on this information, texture information corresponding to the topographic features is extracted from the single-band image. For example, in the red single-band image of region A, region 1 is blocky, with small undulations and a flat shape, and the "smooth" texture is extracted; region 2 is long and strip-shaped, with large undulations and a steep shape, and the "rough" texture is extracted.

[0167] For example, when processing the multispectral image of the target in region A, step d1 splits the image into two single-band images based on the ranges of red light (600-700 nm) and green light (500-600 nm); step d2 calculates the values ​​of adjacent pixels in the red single-band image as 80 and 50, with a difference of 30, a preset threshold of 20 (30 > 20), determines the boundary position, and divides the image into region 1 (pixel value 70-80, concentrated distribution) and region 2 (pixel value 30-50, dispersed distribution); step d3 determines the spatial distribution pattern of the red single-band image as "concentrated distribution + dispersed distribution" based on the distribution characteristics; step d4 analyzes that region 1 is blocky, with small undulations and a flat shape, and extracts the "smooth" texture; region 2 is long and strip-shaped, with large undulations and a steep shape, and extracts the "rough" texture.

[0168] By executing steps c1 to c4, this embodiment of the application splits multispectral images into single-band images, providing a basis for targeted analysis of terrain features at different wavelengths; calculates pixel differences and divides regions to accurately locate boundaries with significant terrain changes; summarizes spatial distribution patterns and extracts the overall rules of pixel arrangement; and finally, the extracted texture information can accurately reflect the surface features of the terrain (such as smoothness and roughness), providing realistic surface texture data for subsequent terrain modeling and improving the accuracy of terrain feature description.

[0169] In one possible embodiment, S14, the vegetation index distribution of the topographic area to be measured is dynamically correlated with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of vegetation cover area topography and crop type, including:

[0170] Step 141: Select images of specific bands from the target multispectral image, and obtain the vegetation index distribution of the terrain area to be measured by calculating the proportional relationship of the pixel values ​​corresponding to vegetation in the single-band images of the specific band.

[0171] Among them, the target multispectral image is a topographic image containing information on multiple light wavelengths; the image of a specific band is an image selected from the target multispectral image that corresponds to the light wavelength that shows a clear response to vegetation; the pixel value corresponding to vegetation is the pixel value of the vegetation area in the image of the specific band; the ratio is the ratio of the vegetation pixel values ​​in different images of the specific band; the vegetation index distribution is a numerical distribution reflecting the vegetation growth in the topographic area to be measured, calculated through the ratio; and the result generated after processing is the vegetation index distribution of the topographic area to be measured.

[0172] In this embodiment, firstly, specific band images (such as red and near-infrared bands) that show a significant response to vegetation are selected from the target multispectral image. Then, the proportional relationship of the pixel values ​​corresponding to vegetation in these specific band images is calculated. Specifically, the vegetation pixel value in the near-infrared band is subtracted from the vegetation pixel value in the red band, and then divided by the sum of the vegetation pixel values ​​in these two bands. Finally, the vegetation index distribution of the terrain area to be measured is obtained based on the calculation results. For example, in area A, red and near-infrared band images are selected. At a certain location, the red pixel value is 20 and the near-infrared pixel value is 80. The proportional relationship is (80-20)÷(80+20)=60÷100=0.6. The vegetation index at this location is 0.6. The vegetation index distribution of the entire area is obtained in this way.

[0173] Step 142: Extract elevation information from the fusion features, and determine the range of elevation value variation in the terrain area to be measured based on the elevation information.

[0174] Among them, the fusion feature is the terrain information that integrates the features of multispectral imagery and point cloud data; the elevation information is the height value of each location in the terrain; the range of elevation values ​​is the height difference between the highest and lowest points in the terrain area to be measured; and the result generated after processing is the range of elevation values ​​in the terrain area to be measured.

[0175] In this embodiment of the application, the elevation information (i.e., height value) of each location is first extracted from the fusion features. Then, the maximum and minimum values ​​of these elevation information are found. Finally, the range of elevation value variation is obtained by subtracting the minimum value from the maximum value. For example, in region A, the maximum value of the elevation information extracted from the fusion features is 400 meters, the minimum value is 150 meters, and the range of elevation value variation is 400-150=250 meters.

[0176] Step 143: Correlate the vegetation index distribution with the elevation information according to spatial location to obtain the degree of correlation between the vegetation index distribution value and the elevation value at each spatial location.

[0177] Among them, vegetation index distribution is the numerical distribution reflecting vegetation growth; elevation information is the height value of each location on the terrain; spatial location is the specific location of each point in the terrain; correlation degree is the degree of connection between vegetation index value and elevation value at the same spatial location; the result generated after processing is the correlation degree between vegetation index and elevation at each spatial location.

[0178] In this embodiment, the vegetation index value and corresponding elevation value of each spatial location are first determined. Then, the correlation between these two values ​​is calculated (the correlation is high when the trends of the values ​​are consistent, and low when the trends are opposite). Finally, the correlation of each spatial location is recorded. For example, in area A, the vegetation index at location B is 0.5 and the elevation is 200 meters, while the vegetation index at location C is 0.75 and the elevation is 350 meters. Both values ​​increase with the elevation, and the correlation is 0.8. At location D, the vegetation index is 0.3 and the elevation is 250 meters. The values ​​do not change, and the correlation is 0.4.

[0179] Step 144: Adjust the correlation degree according to the range of elevation value changes, and based on the adjusted correlation degree, combine the growth pattern of crop type under elevation information and vegetation index distribution to generate spatiotemporal dynamic mapping results.

[0180] Among them, the range of elevation values ​​represents the overall undulation of the terrain; the degree of correlation represents the closeness of the connection between vegetation index and elevation; crop type represents the types of plants planted on the terrain; growth pattern represents the growth characteristics of different crops at different heights and vegetation indices; the spatiotemporal dynamic mapping result reflects the correspondence between the terrain and crop type of the vegetation-covered area over time; and the result generated after processing is the spatiotemporal dynamic mapping result.

[0181] In this embodiment, the correlation degree is first adjusted according to the range of elevation value changes (the correlation degree is increased in areas with large elevation changes and decreased in areas with small changes). Then, the growth patterns of different crops are combined (e.g., a certain crop grows well at high altitudes and with high vegetation indices). Finally, the correlation results at different time points are recorded to generate the correspondence between the vegetation cover area terrain and crop type over time, i.e., the spatiotemporal dynamic mapping result. For example, in area A, the correlation degree of mountainous areas with large elevation changes is adjusted to 0.9. In spring, a certain location (elevation 300 meters, index 0.8) corresponds to crop B. In summer, the index of that location becomes 0.9, still corresponding to crop B.

[0182] For example, in the processing of region A, step 141 selects red and near-infrared band images, calculates the vegetation index at location B (red pixel value 30, near-infrared pixel value 90, vegetation index (90-30) ÷ (90+30) = 60 ÷ 120 = 0.5); at location C (red pixel value 10, near-infrared pixel value 70, vegetation index (70-10) ÷ (70+10) = 60 ÷ 80 = 0.75), thus obtaining the vegetation index distribution. Step 142 extracts elevation information from the fused features, with a maximum value of 400 meters and a minimum value of 150 meters, and an elevation variation range of 400-150=250 meters. In step 143, the values ​​at locations B (0.5, 200 meters) and C (0.75, 350 meters) both increase with increasing location, with a correlation degree of 0.8; location D (0.6, 220 meters) has a higher elevation and higher index than B, with a correlation degree of 0.7. In step 144, the elevation of the mountainous area in region A changes by 150 meters, and the correlation degree is adjusted to 0.9; the elevation of the plain changes by 50 meters, and the correlation degree is adjusted to 0.6; combined with the pattern that crop B grows well at high altitude and high index, the position of B in spring corresponds to crop B, and the index of this position in summer becomes 0.6, still corresponding to crop B, thus generating a spatiotemporal dynamic mapping result.

[0183] By executing steps 141 to 144, this embodiment of the application calculates the vegetation index distribution to reflect vegetation growth, determines the elevation change range to reflect terrain undulation, establishes and dynamically adjusts the correlation between vegetation and elevation, and finally generates a spatiotemporal dynamic mapping result by combining crop growth patterns. This makes the terrain model not only contain terrain elevation information, but also incorporate the dynamic change characteristics of vegetation and crops, providing a more comprehensive and realistic reference for flight simulation terrain modeling and path planning.

[0184] Figure 2 A schematic diagram of a multispectral fusion flight simulation terrain generation system provided in this application embodiment is shown below. Figure 2 As shown, the system includes:

[0185] The acquisition module 21 is used to acquire the original multispectral images and original synthetic aperture radar data of the terrain area to be measured, identify the cloud coverage rate based on the pixel value distribution of the original multispectral images, and filter the target multispectral images and target synthetic aperture radar data of the corresponding time window with cloud coverage rates lower than a set threshold.

[0186] Extraction module 22 is used to extract terrain edge features from target multispectral images and dynamically adjust the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density of complex terrain areas, including steep slopes and canyon areas, is increased by increasing the tilt of the radar beam incident angle.

[0187] Module 23 is used to build a dual-branch processing model. The first branch processes the target multispectral image, and the second branch processes the point cloud data. The high-resolution terrain features output by the two branches are fused through a weighted mechanism to obtain the fused features.

[0188] The association module 24 is used to dynamically associate the vegetation index distribution of the terrain area to be measured with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of the vegetation cover area terrain and crop type.

[0189] The generation module 25 is used to generate a three-dimensional flight simulation terrain model with semantic labels based on the spatiotemporal dynamic mapping results, for the planning of flight simulation paths.

[0190] Figure 2 The aforementioned multispectral fusion flight simulation terrain generation system can perform... Figure 1 The implementation principle and technical effects of the multispectral fusion flight simulation terrain generation method described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the multispectral fusion flight simulation terrain generation system in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0191] In one possible design, Figure 2 The multispectral fusion flight simulation terrain generation system of the illustrated embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32.

[0192] The storage component 31 stores one or more computer instructions, which are invoked and executed by the processing component 32. The processing component 32 is used to execute the specific process of the above-described multispectral fusion flight simulation terrain generation method.

[0193] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0194] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Random Access Memory (RAM), Static Random-Access Memory (SRAM), Erasable Programmable Read Only Memory (EPROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0195] Of course, computing devices may also include other components, such as input / output interfaces, display components, and communication components. Input / output interfaces provide an interface between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between the computing device and other devices.

[0196] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0197] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The illustrated embodiment is a multispectral fusion method for generating flight simulation terrain.

[0198] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0199] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0200] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0201] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating flight simulation terrain through multispectral fusion, characterized in that, include: Collect raw multispectral images and raw synthetic aperture radar data of the terrain area to be measured. Identify cloud coverage based on the pixel value distribution of the raw multispectral images. Filter target multispectral images and target synthetic aperture radar data of the corresponding time window with cloud coverage below a set threshold. Topographic edge features are extracted from the target multispectral image, and the radar beam incident angle is dynamically adjusted based on the topographic edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density of the complex terrain area is increased by increasing the tilt of the radar beam incident angle. The complex terrain area includes steep slopes and canyon areas. A dual-branch processing model is constructed, with the first branch processing the target multispectral image and the second branch processing the point cloud data. The high-resolution terrain features output from the two branches are fused through a weighted mechanism to obtain the fused features. The vegetation index distribution of the terrain area to be measured is dynamically correlated with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of vegetation cover area terrain and crop type. Based on the spatiotemporal dynamic mapping results, a three-dimensional flight simulation terrain model with semantic tags is generated for flight simulation path planning; The method of dynamically adjusting the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution includes increasing the point cloud density in complex terrain areas by increasing the tilt of the radar beam incident angle during the generation process. By analyzing the terrain edge features, the outline of the terrain region to be measured is generated, the curvature change rate of the outline is calculated, and terrain edge feature lines are generated based on the curvature change rate. Based on the spatial distribution pattern of the terrain edge feature lines, the terrain area to be measured is divided into multiple sub-regions with similar terrain features, and the corresponding dominant slope aspect of each sub-region is determined. Based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region, the complexity value of each sub-region is calculated, and the radar beam incident angle is adjusted according to the complexity value. The control radar scans each of the sub-regions according to the adjusted radar beam incident angle, obtains the reflection signal of each of the sub-regions, analyzes the return time and intensity of the reflection signal, and determines the position of the first reflection point. An initial point cloud data set is generated based on the location of the first reflection point. In the initial point cloud data, for areas with complex terrain, the point cloud density of the complex terrain area is increased by increasing the tilt of the radar beam incident angle, so as to generate point cloud data with non-uniform resolution.

2. The multispectral fusion method for generating flight simulation terrain according to claim 1, characterized in that, The calculation of the complexity value of each sub-region based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region includes: The total length of all the terrain edge feature lines in each of the sub-regions is calculated, and the total length is divided by the area of ​​the corresponding sub-region to obtain the density value of the terrain edge feature lines in the unit sub-region. Analyze the range of the angle between the dominant slope aspect and the due north direction in each of the sub-regions, and divide the range of the angle into multiple angle intervals; Determine the corresponding slope aspect coefficient based on the included angle range of the dominant slope aspect of each sub-region; Calculate the average curvature of the terrain edge feature lines in each sub-region, and divide the average curvature into multiple levels. Each level of the average curvature corresponds to a different complexity correction factor. The product of the density value of the terrain edge feature line in the unit sub-region, the corresponding slope aspect coefficient, and the corresponding complexity correction factor is used as the complexity value of each sub-region.

3. The multispectral fusion method for generating flight simulation terrain according to claim 1, characterized in that, The step of generating an initial point cloud data set based on the position of the first reflection point, and increasing the point cloud density in areas with complex terrain by increasing the tilt of the radar beam incident angle to generate point cloud data with non-uniform resolution, includes: Arrange the position information of all the first reflection points according to three-dimensional spatial coordinates to generate an initial point cloud data set; Each initial point cloud data in the initial point cloud dataset is assigned to a corresponding sub-region; Extract the sub-regions that are considered to be complex terrain regions from all sub-regions that have been allocated initial point cloud data, and control the target synthetic aperture radar to scan the complex terrain regions multiple times according to the target radar beam incident angle tilt to obtain the location information of the second reflection point. The target radar beam incident angle tilt is greater than the adjusted radar beam incident angle. The location information of the second reflection point is added to the initial point cloud data of the corresponding complex terrain area to obtain the target point cloud data of the complex terrain area, so as to improve the point cloud density of the complex terrain area. By comparing the point cloud density corresponding to each of the sub-regions, it is verified that the point cloud density of the terrain-complex region is higher than that of the non-terrain-complex region in the terrain region to be tested. After the verification is passed, point cloud data with non-uniform resolution is generated based on the target point cloud data of the terrain-complex region and the initial point cloud data of the non-terrain-complex region.

4. The multispectral fusion method for generating flight simulation terrain according to claim 1, characterized in that, The high-resolution terrain features include texture information and geometric information; The constructed dual-branch processing model involves the first branch processing the target multispectral image and the second branch processing the point cloud data. A weighted mechanism is used to fuse the high-resolution terrain features output from both branches to obtain fused features, including: Based on the first branch, the target multispectral image is decomposed into single-band images corresponding to multiple bands, the spatial distribution pattern of pixel values ​​in each single-band image is determined, and texture information corresponding to terrain features is extracted based on the spatial distribution pattern. Based on the second branch, the point cloud data is organized into multiple data groups according to spatial location relationships. The distribution density and relative location relationships of the point cloud data in each data group are analyzed, and geometric information corresponding to terrain features is extracted. The confidence weight of the texture information is determined based on the acquisition parameters of the target multispectral image, and the confidence weight of the geometric information is determined based on the acquisition parameters of the point cloud data. The texture information and the geometric information are weighted and merged according to their corresponding confidence weights to obtain the fused feature.

5. The multispectral fusion method for generating flight simulation terrain according to claim 4, characterized in that, The step of decomposing the target multispectral image into multiple single-band images corresponding to each band based on the first branch, determining the spatial distribution pattern of pixel values ​​in each single-band image, and extracting texture information corresponding to terrain features based on the spatial distribution pattern includes: The target multispectral image is divided into multiple independent single-band images according to a preset light wavelength range; Based on the values ​​of adjacent pixels in each single-band image, determine the positions where the value difference between adjacent pixels in each single-band image is greater than a preset difference threshold. According to the positions where the value difference is greater than the preset difference threshold, divide the corresponding single-band image and record the distribution characteristics of each region in the single-band image. Based on the distribution characteristics of each region, determine the spatial distribution pattern of pixel values ​​in each single-band image; Based on the shape, topographic relief, and topographic morphology of each region in the spatial distribution pattern, texture information corresponding to the topographic features is extracted from the single-band image.

6. The multispectral fusion method for generating flight simulation terrain according to claim 1, characterized in that, The step of dynamically associating the vegetation index distribution of the topographic region to be measured with the elevation information in the fused features to generate a spatiotemporal dynamic mapping result of vegetation cover topography and crop type includes: Images of specific bands are selected from the target multispectral image, and the vegetation index distribution of the terrain area to be measured is obtained by calculating the proportional relationship of the pixel values ​​corresponding to vegetation in the single-band images of the specific bands. Elevation information is extracted from the fused features, and the range of elevation values ​​in the terrain region to be measured is determined based on the elevation information; The vegetation index distribution and the elevation information are correlated according to spatial location to obtain the degree of correlation between the vegetation index distribution value and the elevation value at each spatial location. Based on the range of changes in the elevation values, the correlation degree is adjusted, and based on the adjusted correlation degree, combined with the growth patterns of crop types under the elevation information and the vegetation index distribution, a spatiotemporal dynamic mapping result is generated.

7. A multispectral fusion flight simulation terrain generation system, characterized in that, include: The acquisition module is used to acquire raw multispectral images and raw synthetic aperture radar data of the terrain area to be measured, identify cloud coverage based on the pixel value distribution of the raw multispectral images, and filter target multispectral images and target synthetic aperture radar data with cloud coverage below a set threshold and corresponding time windows. The extraction module is used to extract terrain edge features from the target multispectral image and dynamically adjust the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution. During the generation process, the point cloud density of the terrain complex area is increased by increasing the tilt of the radar beam incident angle. The terrain complex area includes steep slopes and canyon areas. The module is used to build a dual-branch processing model. The first branch processes the target multispectral image, and the second branch processes the point cloud data. The high-resolution terrain features output by the two branches are fused through a weighted mechanism to obtain the fused features. The association module is used to dynamically associate the vegetation index distribution of the terrain area to be measured with the elevation information in the fusion features to generate a spatiotemporal dynamic mapping result of vegetation cover area terrain and crop type. The generation module is used to generate a three-dimensional flight simulation terrain model with semantic tags based on the spatiotemporal dynamic mapping results, for the planning of flight simulation paths; The method of dynamically adjusting the radar beam incident angle based on the terrain edge features to generate point cloud data with non-uniform resolution includes increasing the point cloud density in complex terrain areas by increasing the tilt of the radar beam incident angle during the generation process. By analyzing the terrain edge features, the outline of the terrain region to be measured is generated, the curvature change rate of the outline is calculated, and terrain edge feature lines are generated based on the curvature change rate. Based on the spatial distribution pattern of the terrain edge feature lines, the terrain area to be measured is divided into multiple sub-regions with similar terrain features, and the corresponding dominant slope aspect of each sub-region is determined. Based on the density value of the terrain edge feature lines and the corresponding dominant slope aspect of each sub-region, the complexity value of each sub-region is calculated, and the radar beam incident angle is adjusted according to the complexity value. The control radar scans each of the sub-regions according to the adjusted radar beam incident angle, obtains the reflection signal of each of the sub-regions, analyzes the return time and intensity of the reflection signal, and determines the position of the first reflection point. An initial point cloud data set is generated based on the location of the first reflection point. In the initial point cloud data, for areas with complex terrain, the point cloud density of the complex terrain area is increased by increasing the tilt of the radar beam incident angle, so as to generate point cloud data with non-uniform resolution.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a multispectral fusion flight simulation terrain generation method as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a multispectral fusion method for generating flight simulation terrain as described in any one of claims 1-6.

Images