A space-air-ground data fusion and spatial layout optimization method for rural planning

Abstract

The present application relates to the technical field of multi-source data fusion processing, and more particularly to a space-air-ground data fusion and spatial layout optimization method for rural planning, which comprises the following steps: calculating local geometric features of each pixel, which are positively correlated with the variance of elevation values in the window neighborhood of the pixel and the modulus of the gradient of the elevation value of the center point of the window neighborhood; constructing spectral reliability of the pixel based on the local geometric features; constructing fusion features of the pixel by weighting based on the spectral reliability; performing semantic segmentation on the rural area based on the fusion features of the pixel to obtain rural feature types, and combining the feature types with a digital elevation model to perform iterative optimization on the rural planning layout by using a preset optimization algorithm, thereby effectively improving the accuracy of the optimization result.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion processing technology. More specifically, this invention relates to a method for air-space-ground data fusion and spatial layout optimization for rural planning. Background Technology

[0002] The core of scientific rural planning lies in the precise allocation of natural resources and settlements. Currently, the basic data for planning mainly relies on space-based satellite remote sensing and airborne drone aerial surveys. Satellite imagery has wide coverage but limited spatial resolution, often resulting in the inability to distinguish subtle features such as narrow paths in fields due to pixel mixing; while drone low-altitude remote sensing has extremely high resolution, it is mostly in the visible light band and lacks spectral characteristics, making it difficult to distinguish similar features such as lawns and green corrugated iron roofs.

[0003] To overcome the limitations of a single data source, existing technologies typically employ air-space-ground data fusion algorithms, which mainly combine the multispectral advantages of satellites with the high-texture advantages of UAVs through geometric registration and channel stitching.

[0004] However, in practical applications, existing fusion algorithms often employ globally consistent registration logic, neglecting the geometric differences in aerospace perspectives caused by complex rural terrain. Because the projection geometry of satellite high-altitude vertical observations and UAV low-altitude perspectives is inconsistent, existing fusion algorithms cannot adaptively adjust their fusion logic based on terrain geometry when there are surface undulations or abrupt changes in building height. This deficiency leads to severe edge ghosting and spectral misalignment in synthetic images at terrain fracture zones or building edges, not only reducing data quality but also potentially causing planning schemes generated based on this data to be out of sync with the actual terrain, and even posing a risk of planning construction land in areas prone to geological hazards.

[0005] Therefore, how to adaptively adjust the fusion logic based on terrain geometric features to improve data quality and the accuracy of layout optimization results is an urgent problem to be solved. Summary of the Invention

[0006] To address the aforementioned technical challenge of how to adaptively adjust the fusion logic based on terrain geometric features to improve data quality and the accuracy of layout optimization results, this invention proposes a method for air-space-ground data fusion and spatial layout optimization for rural planning. This method includes the following steps:

[0007] This process involves acquiring pixels from satellite imagery, visible light orthophotos, and digital elevation models (DEMs) containing multispectral features in rural areas. Local geometric features of each pixel are calculated; these features are positively correlated with the variance of elevation values ​​within the pixel's neighborhood and with the magnitude of the elevation gradient at the center point of the neighborhood. Spectral confidence of pixels is constructed based on these local geometric features; it is negatively correlated with the local geometric features, positively correlated with the pixel's brightness value in the visible light orthophotos, and negatively correlated with the mean brightness of all visible light orthophoto pixels within the satellite image pixel's range. A fusion feature is constructed for each pixel, including pixel edge intensity weighted by subtracting spectral confidence from 1, and satellite image spectral features weighted by spectral confidence. Based on the fusion feature, semantic segmentation of rural areas is performed to obtain rural land cover types. Combining these land cover types with the DEM, a pre-defined optimization algorithm is used to iteratively optimize the rural planning layout.

[0008] This invention provides a method for air-space-ground data fusion and spatial layout optimization for rural planning, which can improve the accuracy of optimization results. In the optimization process, existing air-space-ground data fusion technologies typically employ globally consistent registration logic, neglecting the geometric differences in projection from the air-space perspective caused by complex rural terrain, resulting in edge ghosting and spectral misalignment. This invention assesses terrain complexity and projection deviation risk by calculating elevation variance and gradient modulus, constructs a spectral reliability index, and dynamically adjusts the fusion weights of satellite spectral features and UAV edge intensity accordingly. In flat terrain areas, rich satellite spectral information is retained to distinguish ground material textures, while in areas with large terrain undulations, high-definition edges from UAVs are used to lock boundaries, effectively solving the problems of mixed pixel interference and geometric distortion, improving semantic segmentation accuracy, and, based on this, achieving scientific and safe iterative optimization of rural spatial layout using a digital elevation model.

[0009] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning includes the following steps: acquiring pixels in rural areas from satellite images, visible light orthophotos, and digital elevation models containing multispectral features; acquiring data sources from multispectral satellite images, visible light orthophotos, and digital elevation models respectively; mapping all data sources to the CS2000 geodetic coordinate system; upsampling the satellite images using bicubic interpolation based on the resolution of the visible light orthophotos to obtain satellite images to be fused with the same resolution as the visible light orthophotos, thus achieving spatial pixel-level alignment of the three data sources; and obtaining the spectral features corresponding to each pixel in the satellite images, the brightness value corresponding to the visible light orthophotos, and the elevation value corresponding to the digital elevation model after preprocessing.

[0010] This invention achieves strict pixel-level alignment of multi-source heterogeneous data at the physical level through a unified spatial reference mapping and bicubic interpolation upsampling based on the resolution of visible light orthophotos. This eliminates physical offsets between different data sources, providing a spatially consistent data foundation for subsequent feature extraction.

[0011] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning includes a method for acquiring the neighborhood of a cell window, comprising: constructing a grid with a side length of [missing information] centered on the current cell. The window neighborhood, where, It is an odd number; the center point of the window neighborhood is the current cell.

[0012] This invention provides a method for constructing a cell's window neighborhood. By defining a window neighborhood with odd-numbered side lengths, the calculation range of local geometric features is clearly defined. This centrally symmetric window design ensures that the terrain fluctuations of the surrounding environment are fully considered when calculating the elevation variance and gradient of the current cell.

[0013] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning includes calculating the local geometric features of each pixel, comprising:

[0014] ;

[0015] For pixels Local geometric features; This represents the number of pixels within the window's neighborhood. For pixels The window neighborhood; For window neighborhood The elevation value of the i-th pixel within the range; For window neighborhood The average elevation value of all pixels within the area; To prevent division by zero; For pixels elevation gradient, This is the modulo symbol.

[0016] This invention provides a precise method for calculating local geometric features. By combining the variance of elevation values ​​within a window to reflect the severity of undulations and the gradient magnitude to reflect the rate of slope change, it can accurately identify high-risk areas such as steep slopes and the edges of residential land that are prone to nonlinear projection deviations.

[0017] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning includes, wherein the spectral reliability of pixels is constructed based on local geometric features, comprising:

[0018] ;

[0019] , , Each pixel Spectral reliability, local geometric features, and brightness values ​​in visible light orthophotos; This is a terrain fragmentation adjustment factor; , These represent the maximum and minimum elevation values ​​within the rural area, respectively. The elevation value is expressed in units. For rural areas, the pixel range of satellite imagery Brightness of all visible light orthophoto pixels Mean; To determine the sign of the mean, For An exponential function with base 0.

[0020] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning includes obtaining the edge intensity of each pixel by using the Canny operator for edge detection.

[0021] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning is provided. The method combines land cover types and digital elevation models, and uses a preset optimization algorithm to iteratively optimize the rural planning layout. The method includes: semantically segmenting the rural area based on fusion features to obtain multiple grids; calculating the layout suitability score of each grid, wherein the suitability score is positively correlated with the land cover type score extracted based on fusion features and negatively correlated with the slope extracted based on the digital elevation model; and adjusting the land use type of each grid using the optimization algorithm to maximize the total suitability score, thereby obtaining the final rural spatial layout scheme.

[0022] This invention provides a method for calculating the overall suitability score. By combining land feature type attributes with terrain slope constraints, it uses an optimization algorithm to automatically find the optimal layout, solving the problem that traditional planning relies solely on experience and ignores micro-topographic risks. This ensures that the planning scheme meets land use function requirements while avoiding steep slope areas unsuitable for construction.

[0023] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning is provided. The calculation of the layout suitability score of each grid includes: weighting the land feature type score and slope of each grid based on a preset weight to obtain the layout suitability score.

[0024] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning is provided, wherein the preset optimization algorithm is a genetic algorithm.

[0025] According to the present invention, a method for air-space-ground data fusion and spatial layout optimization for rural planning is provided. The optimization algorithm further includes a hard constraint: prohibiting the planning of geological disaster risk areas as residential land.

[0026] The present invention has the following beneficial effects:

[0027] Based on the above technical solutions, this invention provides a method for air-space-ground data fusion and spatial layout optimization for rural planning. It assesses terrain complexity and projection deviation risks by calculating elevation variance and gradient modulus, constructs a spectral reliability index, and dynamically adjusts the fusion weights of satellite spectral features and UAV edge intensity accordingly. In flat terrain areas, it retains rich satellite spectral information to distinguish ground features, while in areas with large terrain undulations, it utilizes high-resolution UAV edges to lock boundaries, effectively solving the problems of mixed pixel interference and geometric distortion, improving semantic segmentation accuracy, and, based on this, combining with a digital elevation model to achieve scientific and safe iterative optimization of rural spatial layout. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the steps of a method for air-space-ground data fusion and spatial layout optimization for rural planning according to an embodiment of the present invention.

[0029] Figure 2 A schematic diagram of the spectral reliability distribution of a pixel provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram illustrating the data fusion effect of a prior art embodiment provided by the present invention;

[0031] Figure 4 This is a schematic diagram illustrating the data fusion effect of an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0033] Please see Figure 1 The diagram illustrates a flowchart of a method for air-space-ground data fusion and spatial layout optimization for rural planning, provided by an embodiment of the present invention. The method includes the following steps:

[0034] S1: Acquire satellite imagery, visible light orthophotos, and pixels from digital elevation models that contain multispectral features in rural areas.

[0035] Specifically, multispectral satellite imagery of rural areas is acquired from a space-based platform; visible light orthophotos of rural areas are acquired using a low-altitude UAV platform; and a digital elevation model is generated through photogrammetry. The rural area can be any region within a rural area, which is also the area requiring optimization in this embodiment of the invention.

[0036] It is important to note that the acquisition and preprocessing of multi-source heterogeneous data is the physical and logical starting point for the space-air-ground data fusion system. In the actual scenario of rural planning, space-based satellites, airborne drones, and geographic information models are at completely different observation altitudes, imaging systems, and spatial resolution levels. Without physical spatial benchmark alignment and pixel-level resampling, all subsequent feature extraction and fusion calculations will lose their geographical meaning. Otherwise, pixels from different data sources will experience severe physical offsets, leading to semantic drift during classification. This means that spectral features that should belong to farmland will be incorrectly projected onto the coordinates of rural settlements, resulting in fundamental errors in the planning base map.

[0037] Furthermore, rural features have varying scales. For example, micro-features such as narrow field roads and irrigation ditches often appear as mixed pixels in low-resolution satellite imagery and cannot be effectively identified. While UAV imagery has extremely high spatial resolution, it is difficult to distinguish features with similar materials, such as green lawns and green corrugated iron roofs, due to its single spectral band and lack of infrared and other spectral features.

[0038] Based on this, embodiments of the present invention can establish a unified spatial benchmark and perform high-precision resampling to ensure that each physical coordinate point has a completely consistent mapping relationship in data layers of different dimensions, and provide a spatially consistent data base for subsequent geometric feature correction through physical alignment.

[0039] It should be understood that since the spatial resolution of UAV imagery is much higher than that of satellite imagery, the embodiments of the present invention can use high-resolution visible light orthophotos as a reference to map all acquired satellite imagery, UAV imagery, and digital elevation models to ensure that the same geographic point in the physical world has the same spatial attributes in different data layers.

[0040] For example, in this embodiment of the invention, acquiring pixels in a rural area containing multispectral features from satellite imagery, visible light orthophotos, and a digital elevation model includes: acquiring data sources from multispectral satellite imagery, visible light orthophotos, and the digital elevation model, respectively; mapping all data sources to the CS2000 geodetic coordinate system; using the resolution of the visible light orthophotos as a reference, upsampling the satellite imagery using bicubic interpolation to obtain a satellite image to be fused with the same resolution as the visible light orthophotos, thus achieving spatial pixel-level alignment of the three data sources; and obtaining, after preprocessing, the spectral features corresponding to each pixel in the satellite imagery, the brightness value corresponding to the visible light orthophotos, and the elevation value corresponding to the digital elevation model.

[0041] Preprocessing can be a process of standardizing the data, and the specific settings can be configured according to actual needs.

[0042] After processing using the above steps, the coordinates of each pixel simultaneously point to the texture information in the visible light orthophoto, the elevation information in the digital elevation model, and the multispectral information in the satellite image.

[0043] S2: Calculate the local geometric features of each pixel. The local geometric features are positively correlated with the variance of the elevation values ​​in the neighborhood of the pixel window and positively correlated with the magnitude of the gradient of the elevation value at the center point of the neighborhood.

[0044] It is important to note that the complex and undulating terrain of rural areas is the core cause of geometric discrepancies in the aerospace perspective. Vertical observations from satellites and low-altitude tilted or orthogonal views from drones produce different projection differences when facing abrupt terrain changes such as slopes and the edges of residential sites. Without a quantitative assessment of the geometric heterogeneity of the terrain, the fusion algorithm will treat all areas the same, introducing severe edge ghosting and spectral misalignment in areas with fragmented terrain. This geometric error caused by terrain can lead to a disconnect between the generated planning scheme and the actual terrain, and may even result in the incorrect planning of construction land on steep slopes with landslide risks.

[0045] Based on this, this embodiment of the invention assesses the projection deviation risk of each pixel due to terrain undulations by constructing local geometric feature indices. By analyzing the degree of fluctuation in local elevation and slope characteristics, it is possible to accurately identify which areas belong to high-risk zones of geometric uncertainty. In flat farmland areas, the geometric feature values ​​are low, indicating high data alignment accuracy; while on steep slopes or at the edges of houses, the geometric feature values ​​increase, alerting the system to the risk of nonlinear offset. This adaptive identification mechanism can provide a decision-making basis for subsequent injection of spectral information.

[0046] For example, in an embodiment of the present invention, the method for obtaining the neighborhood of a cell window includes: constructing a neighborhood with a side length of centered on the current cell. The window neighborhood, where, The number is odd; the center point of the window's neighborhood is the current cell. Optionally, in an embodiment of the invention, It can be set to 5, which means it contains 25 physical pixels; the specific setting can be adjusted according to actual needs.

[0047] For ease of understanding, embodiments of the present invention are used to process pixels in rural areas. Let's take an example to illustrate.

[0048] For example, in an embodiment of the present invention, the local geometric features of each pixel are calculated using the following relationship:

[0049] ;

[0050] For pixels Local geometric features; This represents the number of pixels within the window's neighborhood. For pixels The window neighborhood; For window neighborhood The elevation value of the i-th pixel within the range; For window neighborhood The average elevation value of all pixels within the area; To prevent division by zero; For pixels elevation gradient, This is the modulo symbol.

[0051] The zero-prevention parameter can optionally be set to 0.001, and the specific setting can be determined according to actual needs.

[0052] In this relation, the first term The variance, representing the elevation value, reflects the degree of terrain undulation. (Due to the window neighborhood...) The larger the squared difference between the elevation value of the i-th pixel and the mean, the more drastic the elevation fluctuation in that neighborhood, and thus the greater the local geometric features, showing a positive correlation.

[0053] gradient term Represents slope characteristics. Represents slope, dimensionless, gradient magnitude This is used to characterize the rate of change of elevation at that point. Since the greater the slope, the greater the imaging displacement deviation between the satellite and the drone, the larger the value of the gradient term, and the greater the corresponding local geometric features.

[0054] It should be understood that the variance of a single elevation value can only reflect height fluctuations. By introducing gradient magnitude, the influence of abrupt slope changes on the geometric distortion of sensor imaging can be further captured.

[0055] In summary, local geometric features are used to assess the risk of geometric distortion of a pixel due to terrain undulations. The larger the local geometric features of a pixel, the greater the likelihood that the pixel is located on the edge of a steep slope; conversely, the smaller the local geometric features, the greater the likelihood that the pixel is located in a flat area.

[0056] Thus, by assessing the projection deviation risk of each pixel point due to terrain undulations, the embodiments of the present invention can accurately obtain the geometric uncertainty of rural areas, evaluate the impact of terrain complexity on remote sensing imaging quality, and lay a foundation for subsequent spectral information input.

[0057] S3: Construct the spectral reliability of pixels based on local geometric features. The spectral reliability is negatively correlated with the local geometric features, positively correlated with the brightness value of the pixel in the visible light orthophoto, and negatively correlated with the mean brightness value of all visible light orthophoto pixels within the satellite image pixel range.

[0058] It should be noted that pixel interference from satellite imagery is a bottleneck for accurate rural identification. Since a single satellite pixel often covers multiple drone pixels, in areas sensitive to geometric differences, the multispectral information provided by satellite is essentially a mixture of various objects and cannot represent the true local texture. Without introducing a reliability assessment through spectral injection, directly injecting satellite spectra into high-resolution imagery can lead to spectral overflow at ground feature boundaries. For example, the green spectrum of farmland might flow into adjacent roads, causing the road width to be incorrectly identified.

[0059] Based on this, this embodiment of the invention constructs a dynamic filter. By calculating spectral confidence, the system can automatically adjust the weight of spectral information based on terrain complexity and texture consistency. In areas with flat terrain and uniform texture, satellite spectra are given extremely high confidence to accurately distinguish materials using their multispectral features; in areas with fragmented terrain or small features, their weight is significantly reduced, thereby effectively suppressing the interference of unreliable spectra on feature boundaries.

[0060] For example, in an embodiment of the present invention, the spectral confidence of a pixel is constructed based on local geometric features, as shown in the following relationship:

[0061] ;

[0062] For pixels The reliability of the spectrum; This is a terrain fragmentation adjustment factor; For pixels Local geometric features; This represents the maximum elevation value within the rural area. This represents the minimum elevation value within the rural area. The elevation value is expressed in units. For pixels Brightness values ​​in visible light orthophotos; For rural areas, the pixel range of satellite imagery Brightness of all visible light orthophoto pixels Mean; To determine the sign of the mean, For An exponential function with base 0.

[0063] The spectral confidence level ranges from 0 to 1. The terrain fragmentation adjustment factor can be set according to the regional terrain standard deviation; in this embodiment, it can be set to 1.2, but can be set according to actual needs. The unit elevation value in this embodiment can be the normalized average elevation value of the rural area, or a value of 1.

[0064] In this relation, the exponent term This demonstrates the inhibitory effect of terrain on credibility. It represents the maximum relative elevation difference within the rural area under study, signifying the vertical depth of landform development. This is due to local geometric features. The larger the value, the higher the risk of geometric deviation. Since the impact of terrain complexity on image quality is non-linear, high-distortion areas can be quickly masked through exponential functions. Therefore, the smaller the negative value of the exponent, the faster the spectral reliability decays, showing a negative correlation.

[0065] fractional terms To maintain texture consistency, due to The greater the deviation from the mean, the greater the possibility of the existence of independent small ground features, and the more difficult it is to match the satellite spectrum, thus reducing the reliability.

[0066] For details, please refer to Figure 2 , Figure 2 This is a schematic diagram of the spectral reliability distribution of a pixel provided in an embodiment of the present invention.

[0067] As shown in the figure, in the edge areas with high terrain sensitivity, the color turns dark green and the value decays rapidly, indicating that the system automatically identifies the spectral information of these areas as unreliable, thus activating a dynamic filter to suppress interference from mixed pixels. In flat areas, the color is light yellow, indicating that the system automatically identifies the spectral information of these areas as more reliable.

[0068] Thus, based on the above steps, the embodiments of the present invention can obtain the spectral reliability of each pixel. By introducing spectral reliability, the weights of spatial geometric information and spectral information can be adaptively allocated, thereby reducing the interference generated by the mixed pixels in the geometrically fragmented zone in a physically logical manner and improving the accuracy of the fusion result.

[0069] S4: Construct the fused features of the pixels, which include pixel edge intensity weighted by 1 minus spectral confidence, and satellite image spectral features weighted by spectral confidence.

[0070] It is important to note that feature space reconstruction is a crucial step in achieving complementary advantages between air, space, and ground data. Rural planning places extremely high demands on the boundaries and textures of land features. Without feature reconstruction, directly using traditional channel stitching often leads to classification oscillations due to differences in the dimensions or spatial misalignments between different features. The final feature tensor serves as the direct input for semantic segmentation, deeply fusing multi-source features after credibility correction. Traditional stitching methods, lacking credibility considerations, often produce false textures at boundaries. Therefore, it is necessary to organically combine the high-resolution edge intensity of UAVs with the material spectral features of satellites.

[0071] Based on this, embodiments of the present invention can guide feature flow through spectral reliability, identify vegetation using spectral data in flat areas, and lock boundaries using high-definition textures in edge areas.

[0072] Optionally, in this embodiment of the invention, the method for obtaining the pixel edge intensity includes: using the Canny operator to perform edge detection to obtain the edge intensity of each pixel.

[0073] For example, in an embodiment of the present invention, the fusion features of a pixel are constructed using the following relational expression:

[0074] ;

[0075] For pixels The fusion characteristics; For pixels The reliability of the spectrum; For pixels Edge strength; For pixels Spectral characteristics.

[0076] Among them, the spectral features are multispectral component vectors. Edge intensity represents the contour of ground features.

[0077] It should be understood that, due to the large spectral features and small edge intensities with different dimensions, the edge intensities and spectral features in the above formulas are data that have been normalized by Z-score or Min-Max.

[0078] In this relationship, if the spectral confidence level is close to 1, the pixel may be located in an area such as flat farmland. The optimized model mainly receives satellite spectral features, thereby using indices such as NDVI to accurately distinguish between crops and weeds. If the spectral confidence level is close to 0, the pixel may be located in an area such as the edge of a building or a narrow road. The model automatically filters out the blurred spectrum and instead relies on the high-definition edge intensity of the drone to ensure centimeter-level boundary accuracy.

[0079] For details, please refer to Figure 3 and Figure 4 , Figure 3 This is a schematic diagram illustrating the data fusion effect of a prior art embodiment provided by the present invention. Figure 4 This is a schematic diagram illustrating the data fusion effect of an embodiment of the present invention.

[0080] Figure 3 In the process, due to the neglect of geometric differences in spatial perspective, direct fusion resulted in obvious blurring, ghosting, and spectral misalignment at the edges of buildings and roads. Figure 4 The technical solution provided by this invention achieves sub-pixel-level alignment by automatically masking erroneous satellite spectra in low-confidence regions and instead relying on the high-resolution edge texture of the UAV. Combined with... Figure 3 and Figure 4 As can be seen, the data fusion results obtained through the technical solution provided by this invention can effectively improve the clarity of rural areas, thus laying a foundation for subsequent layout optimization.

[0081] Thus, by reconstructing the feature tensor, embodiments of the present invention can generate composite feature data that combines high spatial resolution and high spectral resolution.

[0082] S5: Based on the fusion features of pixels, semantic segmentation of rural areas is performed to obtain rural land cover types. Then, by combining land cover types with digital elevation models, a preset optimization algorithm is used to iteratively optimize the rural planning layout.

[0083] For example, when semantic segmentation is used to obtain rural feature types, a deep learning semantic segmentation network can be employed. The deep learning semantic segmentation network can be an algorithm such as U-Net or DeepLabV3+, and the specific configuration can be chosen based on actual needs.

[0084] It should be noted that optimizing rural spatial layout is the final step in the transformation of research results. Traditional planning often involves blindly dividing functional zones without considering micro-topography and dynamic risks. Inaccurate classification may lead to the misclassification of protected forest land as construction land or the planning of residential areas in areas with excessively steep slopes, creating potential geological hazards.

[0085] Based on this, embodiments of the present invention utilize high-precision land feature identification results and terrain models to construct a suitability evaluation function, thereby achieving pixel-level identification of land feature types.

[0086] It should be understood that, in this embodiment of the invention, after processing according to the above steps, each pixel input to the semantic segmentation model is not a single color value, but a multi-dimensional feature vector. For flat farmland or woodland, the spectral confidence is close to 1, and the feature vector contains high-weight satellite multispectral information, which the model can use to accurately distinguish the area. For areas such as the edges of houses or narrow field paths, the spectral confidence is close to 0, and the feature vector is dominated by the edge intensity extracted by the drone, which the model can use to force the segmentation of clear geographical boundaries and avoid edge adhesion.

[0087] Specifically, when implementing semantic segmentation, land features can be divided into construction land, agricultural land, transportation land, and water areas according to the needs of rural planning. For example, construction land includes residential areas, agricultural land includes arable land, orchards, and forest land, and transportation land includes main rural roads and field paths. Corresponding land feature type scores are set for the degree to which each type of land feature is suitable as construction land. The higher the vegetation coverage, the lower the corresponding land feature type score. For example, construction land can be set to 100 points, and basic farmland can be set to 0 points.

[0088] When performing segmentation based on deep learning semantic segmentation networks, digital elevation models (DEMs) are needed to physically constrain the classification results. For example, water surfaces should not appear on steep slopes. If classification results conflict, the categories are revised based on terrain sensitivity geometric features. Finally, morphological processing is performed on the identified features to fill in minor breaks caused by vegetation shading, forming complete vector boundaries that can be directly used for planning and drawing. Specific steps can be implemented using existing technologies, and will not be elaborated upon in this embodiment of the invention.

[0089] The above steps can be used to segment rural land features and divide the rural area into multiple grids. Each grid has a corresponding land feature type score, which can be used to realize rural planning layout.

[0090] For example, in this embodiment of the invention, combining land cover types and digital elevation models, a preset optimization algorithm is used to iteratively optimize the rural planning layout, including: semantically segmenting the rural area based on fusion features to obtain multiple grids, calculating the layout suitability score of each grid, the suitability score being positively correlated with the land cover type score extracted based on fusion features and negatively correlated with the slope extracted based on the digital elevation model; and using the optimization algorithm to adjust the land use type of each grid to maximize the total suitability score, thereby obtaining the final rural spatial layout scheme.

[0091] Optionally, in this embodiment of the invention, the preset optimization algorithm is a genetic algorithm.

[0092] For example, in an embodiment of the present invention, calculating the layout suitability score of each grid includes: weighting the land feature type score and slope of each grid based on a preset weight to obtain the layout suitability score.

[0093] The layout suitability score is used to assess the scientific validity of the current grid as a site for planning; a higher score indicates greater suitability for construction. The feature type score represents the attribute rating of the semantic segmentation results. Slope represents the impact of terrain on construction.

[0094] In mountainous rural areas, the steeper the slope, the higher the construction cost and the lower the safety, resulting in a lower layout suitability score.

[0095] Specifically, due to the high importance of ecological red lines, priority should be given to ecological red lines when optimizing the layout, and the higher the score of land feature type, the more suitable it is for construction. Therefore, the weight of the land feature type score can be greater than the weight of the slope. Optionally, in this embodiment of the invention, the weight of the land feature type score can be set to 0.7, and the weight of the slope can be set to 0.3, which can be set according to actual needs.

[0096] Thus, in optimizing rural layout, this embodiment of the invention calculates the layout suitability score of each grid, which can maximize the adaptation to complex rural terrain features while ensuring that the planning layout meets the ecological bottom line, and cluster the planning layout towards flat, non-ecological areas.

[0097] Optionally, in this embodiment of the invention, the optimization algorithm further includes a hard constraint: prohibiting the planning of geological disaster risk areas as residential land.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for air-space-ground data fusion and spatial layout optimization for rural planning, characterized in that, include: Acquire pixels in rural areas using satellite imagery, visible light orthophotos, and digital elevation models that contain multispectral features; Calculate the local geometric features of each pixel. The local geometric features are positively correlated with the variance of the elevation values ​​in the neighborhood of the pixel window and positively correlated with the magnitude of the gradient of the elevation value at the center point of the neighborhood. The spectral reliability of a pixel is constructed based on local geometric features. The spectral reliability is negatively correlated with the local geometric features, positively correlated with the brightness value of the pixel in the visible light orthophoto, and negatively correlated with the mean brightness value of all visible light orthophoto pixels within the range of the satellite image pixel. Construct fused features for pixels, which include pixel edge intensity weighted by 1 minus spectral confidence, and satellite image spectral features weighted by spectral confidence. Based on the fusion features of pixels, semantic segmentation of rural areas is performed to obtain rural land cover types. Then, by combining land cover types with digital elevation models, a preset optimization algorithm is used to iteratively optimize the rural planning layout.

2. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The acquisition of pixels in rural areas, including satellite imagery with multispectral features, visible light orthophotos, and digital elevation models, includes: Data sources including multispectral satellite imagery, visible light orthophotos, and digital elevation models were collected separately. All data sources were mapped to the CS2000 geodetic coordinate system. Based on the resolution of the visible light orthophotos, bicubic interpolation was used to upsample the satellite imagery to obtain a satellite image to be fused with the same resolution as the visible light orthophotos, thus achieving spatial pixel-level alignment of the three data sources. After preprocessing, the spectral features of each pixel in the satellite image, the brightness value in the visible light orthophoto, and the elevation value in the digital elevation model are obtained.

3. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, Methods for obtaining the neighborhood of a cell window include: Using the current pixel as the center, construct a structure with a side length of... The window neighborhood, where, It is an odd number; the center point of the window neighborhood is the current cell.

4. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The calculation of the local geometric features of each pixel includes: ； For pixels Local geometric features; This represents the number of pixels within the window's neighborhood. For pixels The window neighborhood; For window neighborhood The elevation value of the i-th pixel within the range; For window neighborhood The average elevation value of all pixels within the area; To prevent division by zero; For pixels elevation gradient, This is the modulo symbol.

5. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The spectral reliability of pixels constructed based on local geometric features includes: ； , , Each pixel Spectral reliability, local geometric features, and brightness values ​​in visible light orthophotos; This is a terrain fragmentation adjustment factor; , These represent the maximum and minimum elevation values ​​within the rural area, respectively. The elevation value is expressed in units. For rural areas, the pixel range of satellite imagery Brightness of all visible light orthophoto pixels Mean; To determine the sign of the mean, For An exponential function with base 0.

6. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, Methods for obtaining pixel edge intensity include: The edge intensity of each pixel is obtained by using the Canny operator for edge detection.

7. The method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The method combines land cover types and digital elevation models, and uses a preset optimization algorithm to iteratively optimize the rural planning layout, including: Based on the fusion features, the rural area is semantically segmented to obtain multiple grids. The layout suitability score of each grid is calculated. The suitability score is positively correlated with the land cover type score extracted based on the fusion features and negatively correlated with the slope extracted based on the digital elevation model. An optimization algorithm was used to adjust the land use type of each grid to maximize the overall suitability score, resulting in the final rural spatial layout scheme.

8. A method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 7, characterized in that, The calculation of the layout suitability score for each grid includes: The land cover type score and slope of each grid are weighted according to preset weights to obtain the layout suitability score.

9. A method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The preset optimization algorithm is a genetic algorithm.

10. A method for air-space-ground data fusion and spatial layout optimization for rural planning according to claim 1, characterized in that, The optimization algorithm also includes a hard constraint: prohibiting the planning of geological disaster risk areas as residential land.

Images