Method and system for determining regional boundary of territorial space planning based on remote sensing image

Abstract

The application relates to the technical field of image fusion, in particular to a land space planning regional boundary determination method and system based on remote sensing images. The method comprises the following steps: acquiring target multispectral images and target panchromatic images which are spatially aligned in a target region; performing multi-scale multi-direction transformation on the target panchromatic images and the target multispectral images to obtain a low-frequency subband and a plurality of high-frequency subbands corresponding to each image; fusing each pair of high-frequency subbands based on gradient difference and boundary saliency difference between the high-frequency subbands to obtain a plurality of fused high-frequency subbands; performing multi-scale multi-direction inverse transformation on the plurality of fused high-frequency subbands and the low-frequency subband of the target multispectral images to obtain a target fusion image; and determining a regional boundary based on the target fusion image. The method can obtain a target fusion image with higher resolution and clearer boundaries, and obtain a more accurate and reliable regional boundary.

Description

Technical Field

[0001] This application relates to the field of image fusion technology, specifically to a method and system for determining the boundaries of a land spatial planning area based on remote sensing imagery. Background Technology

[0002] With the steady advancement of my country's territorial spatial planning theory, technology, and practice, establishing long-term, effective, and unified standards for determining regional boundaries is the spatial reference basis for carrying out territorial spatial planning and various statistics on population, socio-economic factors, and other aspects.

[0003] In related technologies, regional boundaries are typically determined by identifying the boundary lines between different ground features in multispectral images (MS).

[0004] However, multispectral remote sensing images have low resolution and insufficient spatial detail, which may lead to blurred edges, making it difficult to identify the boundaries between different land features and to determine the regional boundaries inaccurately. Summary of the Invention

[0005] To address the technical problems of low resolution and insufficient spatial detail in multispectral remote sensing imagery, which may lead to blurred edges and unclear boundaries between different land features, and inaccurate determination of regional boundaries, this application aims to provide a method and system for determining the boundaries of land spatial planning areas based on remote sensing imagery. The specific technical solution adopted is as follows:

[0006] This application provides a method for determining the boundary of a land spatial planning area based on remote sensing imagery, comprising: acquiring a spatially aligned target multispectral image and a target panchromatic image of the target area; performing multi-scale and multi-directional transformations on the target panchromatic image and the target multispectral image to obtain a low-frequency sub-band and multiple high-frequency sub-bands corresponding to each image, wherein the multiple high-frequency sub-bands on the target panchromatic image correspond one-to-one with the multiple high-frequency sub-bands on the target multispectral image; fusing each pair of high-frequency sub-bands based on the gradient difference and boundary saliency difference between each pair of high-frequency sub-bands to obtain multiple fused high-frequency sub-bands; performing a multi-scale and multi-directional inverse transformation based on the multiple fused high-frequency sub-bands and the low-frequency sub-bands of the target multispectral image to obtain a target fused image; and determining the area boundary based on the target fused image.

[0007] Optionally, a high-frequency subband includes high-frequency coefficients of multiple pixels, and pixels in a pair of high-frequency subbands correspond one-to-one. Based on the gradient difference and boundary saliency difference between each pair of high-frequency subbands, each pair of high-frequency subbands is fused to obtain multiple fused high-frequency subbands. This includes: determining the spatial weights of the two pixels in each pair of pixels based on the gradient difference between each pair of pixels in each pair of high-frequency subbands; determining the spatial injection weights of the two pixels in each pair of pixels based on the boundary saliency difference between each pair of pixels in each pair of high-frequency subbands; and fusing the high-frequency coefficients of each pair of pixels based on the spatial weights and spatial injection weights of the two pixels in each pair of pixels to obtain the high-frequency coefficients of each fused pixel in the multiple fused high-frequency subbands.

[0008] Optionally, the high-frequency coefficients of each pair of pixels are fused based on their respective spatial weights and spatial injection weights to obtain the high-frequency coefficients of each fused pixel in multiple fused high-frequency sub-bands. This includes: fusing the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel in the first panchromatic high-frequency sub-band based on their spatial weights and spatial weights, to obtain the spatial fusion coefficient of the first fused pixel in the high-frequency sub-band after the fusion of the first panchromatic high-frequency sub-band and the first multispectral high-frequency sub-band, where the first panchromatic pixel and the first multispectral pixel are a pair of pixels; fusing the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel based on their spatial injection weights and spatial injection weights, to obtain the spatial injection fusion coefficient of the first fused pixel; and fusing the spatial fusion coefficient and the spatial injection fusion coefficient of the first fused pixel to obtain the high-frequency coefficient of the first fused pixel.

[0009] Optionally, fusing the spatial fusion coefficient and the spatial injection fusion coefficient of the first fused pixel to obtain the high-frequency coefficient of the first fused pixel includes: fusing the spatial fusion coefficient and the spatial injection fusion coefficient of the first fused pixel based on a preset dynamic balance parameter to obtain the high-frequency coefficient of the first fused pixel.

[0010] Optionally, determining the spatial weights of the two pixels in each pair of pixels based on the gradient difference between each pair of pixels in each pair of high-frequency subbands includes: determining the magnitude weight of the first panchromatic pixel based on the ratio of the gradient magnitude of the local region where the first panchromatic pixel is located to the total gradient magnitude, wherein the total gradient magnitude is the sum of the gradient magnitudes of the local region where the first multispectral pixel is located and the gradient magnitudes of the local region where the first panchromatic pixel is located; determining the directional consistency weight of the first panchromatic pixel based on the standard deviation of the gradient direction angle of the first panchromatic pixel; determining the spatial weight of the first panchromatic pixel as the product of the directional consistency weight and the magnitude weight; and determining the spatial weight of the first multispectral pixel based on the spatial weight of the first panchromatic pixel, wherein the sum of the spatial weights of the first panchromatic pixel and the first multispectral pixel is 1.

[0011] Optionally, determining the spatial injection weights of the two pixels in each pair of pixels based on the boundary saliency difference of each pair of pixels in each pair of high-frequency subbands includes: performing image segmentation on the target panchromatic image and the target multispectral image to obtain multiple homogeneous regions included in each image; determining the boundary saliency weight of the first panchromatic homogeneous region based on the ratio of the boundary saliency of the first panchromatic homogeneous region where the first panchromatic pixel is located to the total boundary saliency, where the total boundary saliency is the sum of the boundary saliency of the first panchromatic homogeneous region and the boundary saliency of the first multispectral homogeneous region; determining the dynamic weight factor of the first multispectral pixel based on the spectral variance and Moran's index in the first multispectral homogeneous region where the first multispectral pixel is located; determining the spatial injection weight of the first panchromatic pixel by multiplying the dynamic weight factor by the boundary saliency weight; determining the spatial injection weight of the first multispectral pixel based on the spatial injection weight of the first panchromatic pixel, where the sum of the spatial injection weight of the first panchromatic pixel and the spatial injection weight of the first multispectral pixel is 1.

[0012] Optionally, the method further includes: determining the normalized pixel value of each homogeneous region included in the target panchromatic image; and determining the absolute value of the difference between the normalized pixel value in the first panchromatic homogeneous region and the mean of the normalized pixel values ​​of the adjacent homogeneous regions of the first panchromatic homogeneous region as the boundary salience of the first panchromatic homogeneous region.

[0013] Optionally, determining the dynamic weighting factor based on the spectral variance and Moran's index of the first multispectral homogeneous region where the first multispectral pixel is located includes: determining the homogeneity confidence of the first multispectral homogeneous region based on the spectral variance and Moran's index of the first multispectral homogeneous region where the first multispectral pixel is located; and determining the dynamic weighting factor of the first multispectral pixel as the ratio of the homogeneity confidence of the first multispectral homogeneous region to the maximum homogeneity confidence of the homogeneous regions included in the multispectral image.

[0014] Optionally, the method further includes: acquiring the original multispectral image and the original panchromatic image under the target area; converting the original multispectral image and the original panchromatic image into the same coordinate system; resampling the original multispectral image based on the resolution of the original panchromatic image to obtain an intermediate multispectral image with the same size as the original panchromatic image; and aligning the original panchromatic image and the intermediate multispectral image to obtain the target panchromatic image and the target multispectral image.

[0015] This application also provides a system for determining the boundary of a land spatial planning area based on remote sensing imagery. The system includes an image acquisition module, an image processing module, and a boundary determination module. The image acquisition module acquires spatially aligned target multispectral and panchromatic images of the target area. The image processing module performs multi-scale, multi-directional transformations on the target panchromatic and multispectral images to obtain low-frequency sub-bands and multiple high-frequency sub-bands corresponding to each image, with each high-frequency sub-band corresponding one-to-one with the multiple high-frequency sub-bands on the target multispectral image. The image processing module also fuses each pair of high-frequency sub-bands based on gradient differences and boundary saliency differences, obtaining multiple fused high-frequency sub-bands. The image processing module further performs multi-scale, multi-directional inverse transformations based on the multiple fused high-frequency sub-bands and the low-frequency sub-bands of the target multispectral image to obtain a target fused image. The boundary determination module determines the area boundary based on the target fused image.

[0016] This application has the following beneficial effects:

[0017] In this embodiment, by co-processing the target panchromatic image and the target multispectral image at multiple scales, the spatial detail advantage of the panchromatic image and the spectral information advantage of the multispectral image are fully utilized. By fusing each pair of high-frequency sub-bands and the low-frequency sub-band of the target multispectral image based on the gradient difference and boundary significance difference of each pair of high-frequency sub-bands, the spectral characteristics of ground objects can be maintained while improving the resolution. Therefore, based on the target fused image with higher resolution and clearer boundaries, the delineated area boundaries are more accurate and reliable. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1A flowchart illustrating a method for determining the boundary of a land spatial planning area based on remote sensing imagery, provided as an embodiment of this application;

[0020] Figure 2 A flowchart illustrating another method for determining the boundary of a land spatial planning area based on remote sensing imagery, provided as an embodiment of this application;

[0021] Figure 3 A flowchart illustrating another method for determining the boundary of a land spatial planning area based on remote sensing imagery, provided as an embodiment of this application;

[0022] Figure 4 A flowchart illustrating another method for determining the boundary of a land spatial planning area based on remote sensing imagery, provided as an embodiment of this application;

[0023] Figure 5 This is a structural diagram of a land spatial planning area boundary determination system based on remote sensing imagery, provided as an embodiment of this application. Detailed Implementation

[0024] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method and system for determining the boundary of a land spatial planning area based on remote sensing imagery proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0026] With the steady advancement of my country's territorial spatial planning theory, technology, and practice, establishing long-term, effective, and unified standards for determining regional boundaries is crucial for conducting territorial spatial planning and statistical analysis of population, socio-economic, and other aspects. However, when the spatial resolution of remote sensing imagery is insufficient, irregular areas are prone to edge blurring, leading to inaccurate regional segmentation. Low- to medium-resolution imagery struggles to identify small features (such as ditches and scattered buildings), resulting in errors in the fine boundaries of urban development boundaries, ecological protection red lines, and permanent basic farmland.

[0027] Meanwhile, in the process of improving image resolution through image fusion, traditional fusion rules suffer from severe spectral distortion and weak directionality in high-vegetation areas, failing to capture geometric structures. Specifically, existing image fusion algorithms, such as maximum value, local energy, and average value fusion algorithms, all have significant limitations. While the maximum value method can highlight salient features, it easily leads to local overexposure and loss of detail; the local energy method is sensitive to noise, easily mis-fusions high-energy noise points, and its effectiveness is constrained by window parameters; the average value method blurs features by directly averaging pixels and is extremely sensitive to image registration errors. These methods collectively face the inherent contradiction between detail preservation and noise suppression, often leading to decreased segmentation accuracy at feature boundaries, making it difficult to balance the effective fusion of key information with stable image quality.

[0028] The following description, in conjunction with the accompanying drawings, details a specific scheme for a method and system for determining the boundaries of a land spatial planning area based on remote sensing imagery, as provided in this application.

[0029] Please see Figure 1 The diagram illustrates a flowchart of a method for determining the boundary of a land spatial planning area based on remote sensing imagery, provided in one embodiment of this application.

[0030] like Figure 1 As shown, the method for determining the boundary of a land spatial planning area based on remote sensing imagery includes S101-S105.

[0031] S101. Acquire spatially aligned multispectral and panchromatic images of the target region.

[0032] It should be understood that the target multispectral image and the target panchromatic image have the same pixel size and the pixels correspond one-to-one.

[0033] Optionally, the target area is the area whose boundaries are to be defined.

[0034] In one alternative implementation, the original multispectral image and the original panchromatic image under the target area can be acquired; the original multispectral image and the original panchromatic image can be converted into the same coordinate system; the original multispectral image can be resampled based on the resolution of the original panchromatic image to obtain an intermediate multispectral image with the same size as the original panchromatic image; the original panchromatic image and the intermediate multispectral image can be aligned to obtain the target panchromatic image and the target multispectral image.

[0035] Optionally, the original multispectral image of the target area can be obtained from Landsat sensor multispectral remote sensing data, and the original panchromatic image of the target area can be obtained from Gaofen-1 satellite high-resolution panchromatic remote sensing data.

[0036] Optionally, the original multispectral image and the original panchromatic image are images acquired at the same time.

[0037] Optionally, the coordinate systems of the original multispectral and panchromatic images can be converted to a universal transverse mercator (UTM) using the projection conversion tool in ENVI software to ensure that the two images have the same georeferenced frame and eliminate geometric deviations caused by projection differences.

[0038] Optionally, a triple convolution resampling technique can be used to resample the original multispectral image to obtain an intermediate multispectral image with the same size as the original panchromatic image.

[0039] It should be understood that the resampled intermediate multispectral image has the same pixel grid and spatial dimensions as the original panchromatic image.

[0040] In this embodiment of the application, the alignment process includes geometric correction processing and radiation consistency processing.

[0041] Specifically, the geometric correction process is as follows: Principal component analysis (PCA) is performed on the resampled intermediate multispectral image to reduce its dimensionality and generate a multispectral grayscale image. Histogram matching is performed on the original panchromatic image to generate a panchromatic grayscale image. Then, the scale-invariant feature transform (SIFT) algorithm is used to extract feature points from the two grayscale images respectively. For each feature point in the multispectral grayscale image, the Euclidean distance between its 128-dimensional feature vector and the feature point vectors in the panchromatic grayscale image is calculated, and the two closest and second closest distances are found. The nearest neighbor distance ratio method is used for matching and filtering to determine the initial corresponding point pairs. Finally, the affine transformation matrix is ​​calculated based on the initial corresponding point pairs, and the affine transformation matrix is ​​applied to the intermediate multispectral image to output a multispectral image aligned with the original panchromatic image.

[0042] It should be understood that the affine transformation matrix includes rotation, scaling, translation, and shearing parameters.

[0043] Optionally, when the nearest neighbor distance ratio of the nearest neighbor point is less than the set empirical threshold of 0.6, the nearest neighbor point is determined as the initial corresponding point pair of the feature point.

[0044] It should be understood that the initial pair of corresponding points is a pair of pixels.

[0045] Radiometric consistency processing involves using the grayscale histogram of the original panchromatic image as a reference. Histogram matching is performed on each band of the geometrically corrected multispectral image. The reference benchmark is not the global histogram of the original panchromatic image, but rather the statistical characteristics of a large, homogeneous region (such as pure water or bare soil) selected from both images, characterized by stable land cover types. A mapping function is used to adjust the grayscale distribution of each band in the geometrically corrected multispectral image, ensuring comparable radiometric levels for the same land cover in both images while preserving multispectral information to the greatest extent possible. Specifically, for each band of the geometrically corrected multispectral image, its cumulative distribution function and the cumulative distribution function of the original panchromatic image are calculated. By finding a mapping relationship that makes the cumulative distribution functions equal, the pixel values ​​of the geometrically corrected multispectral image are converted into new pixel values, thus obtaining the target multispectral image with matched radiometric characteristics.

[0046] Alternatively, the original panchromatic image can be directly identified as the target panchromatic image.

[0047] It should be understood that the processing of the original panchromatic and original multispectral images eliminates the data mismatch problem caused by differences in sensors and imaging conditions, providing a reliable data foundation for subsequent analysis and effectively avoiding fusion artifacts caused by data quality issues.

[0048] In this embodiment of the application, the target panchromatic image and the target multispectral image can be subjected to max-min normalization processing, which maps all pixel values ​​to the [0,1] interval.

[0049] It should be understood that the "normalized pixel value" involved in this application refers to the value after standardization processing of the original image. Its core purpose is to eliminate the differences in radiation and scale caused by sensor differences, imaging conditions, etc., so that the pixel values ​​are at the same comparable scale, providing a reliable data basis for subsequent pixel-level fusion and comparison. The standardization processing may include, but is not limited to: the radiance value obtained after radiometric calibration, the surface reflectance obtained after atmospheric correction, or linear stretching processing such as maximum-minimum normalization. In all subsequent steps, whenever pixel values ​​are used for calculation and comparison (such as calculating boundary saliency, gradient magnitude, etc.), unless otherwise specified, it refers to the normalized pixel value obtained after the above preprocessing. S102, Perform multi-scale and multi-directional transformation on the target panchromatic image and the target multispectral image to obtain the low-frequency sub-band and multiple high-frequency sub-bands corresponding to each image.

[0050] Among them, multiple high-frequency sub-bands on the target panchromatic image correspond one-to-one with multiple high-frequency sub-bands on the target multispectral image.

[0051] It should be understood that each image corresponds to one low-frequency sub-band, which is used to carry the spectral characteristics of the ground object (i.e., overall brightness and color). These multiple high-frequency sub-bands are multi-scale and multi-directional high-frequency sub-bands. One high-frequency sub-band corresponds to spatial detail information of one scale and one direction. There is a high-frequency sub-band of the target panchromatic image and a high-frequency sub-band of the target multispectral image at one scale and one direction, respectively.

[0052] It should be understood that the "spatial transients" caused by the boundary of the same ground feature appear as strong signals with high signal-to-noise ratio and accurate positioning in panchromatic images, which are effective spatial details to be injected; while in multispectral images, they appear as weak signals with low signal-to-noise ratio and ambiguous positioning, and may be mixed with invalid high-frequency interference caused by noise and artifacts.

[0053] In one implementation, the multi-scale, multi-directional transformation is a non-subsampled contourlet transform (NSCT). The input to the NSCT is a multispectral image and a panchromatic image of the target, and the output is the low-frequency subband and multi-directional high-frequency subband of each image. The NSCT transform achieves multi-scale, multi-directional decomposition by cascading a non-subsampled tower filter bank and a non-subsampled directional filter bank.

[0054] S103. Based on the gradient and boundary saliency of each high-frequency subband in each pair of high-frequency subbands, each pair of high-frequency subbands is fused to obtain multiple fused high-frequency subbands.

[0055] It should be understood that a pair of high-frequency subbands are a pair of corresponding high-frequency subbands, and after fusion, a fused high-frequency subband is obtained.

[0056] Understandably, because the target panchromatic image has high resolution, its high-frequency subbands represent clear, sharp, and accurately positioned spatial details; conversely, because the target multispectral image has low resolution, its high-frequency coefficients are of poor quality, representing blurry, noisy, and poorly positioned spatial details.

[0057] It is understandable that high-frequency subbands include high-frequency coefficients, and fusing each pair of high-frequency subbands is actually fusing the high-frequency coefficients to obtain the high-frequency coefficients of the fused high-frequency subband.

[0058] Alternatively, the absolute value of the difference between gradient magnitudes can be defined as the gradient difference.

[0059] It should be understood that regions with higher gradient magnitudes correspond to edges, contours, and significant textures in the image; regions with lower gradient magnitudes correspond to flat areas and uniform features in the image. When the gradient magnitude of a certain high-frequency sub-band (e.g., the first panchromatic high-frequency sub-band) in a pair of high-frequency sub-bands is higher and the gradient magnitude difference is greater, more features of the first panchromatic high-frequency sub-band can be used in the fusion process.

[0060] Optionally, the absolute value of the difference in boundary significance can be determined as the boundary significance difference.

[0061] Understandably, the boundary saliency of a high-frequency subband is used to characterize the clarity and reliability of the edges between different land cover types within a high-frequency subband. When the boundary saliency of a high-frequency subband (e.g., the first panchromatic high-frequency subband) in a pair of high-frequency subbands is higher and the difference in boundary saliency is larger, more features of the first panchromatic high-frequency subband can be used during the fusion process.

[0062] Therefore, based on the gradient difference and boundary saliency difference between the two high-frequency subbands included in each pair of high-frequency subbands, when fusing a pair of high-frequency subbands, if the details of the high-frequency subband corresponding to the target panchromatic image (hereinafter referred to as the panchromatic high-frequency subband) are reliable, more clear details of the panchromatic high-frequency subband can be fused; if the details of the panchromatic high-frequency subband are unreliable, more high-frequency subbands corresponding to the target multispectral image (hereinafter referred to as the multispectral high-frequency subband) can be fused. Although the details of the multispectral high-frequency subband are not clear enough, the spectral fidelity can be maintained to the greatest extent and serious distortion can be avoided.

[0063] S104. Perform multi-scale, multi-directional inverse transformation based on multiple fused high-frequency sub-bands and low-frequency sub-bands of the target multispectral image to obtain the target fused image.

[0064] It should be understood that the fused target image is a high-resolution multispectral fused image.

[0065] Based on the above description, it should be understood that the low-frequency subband carries the spectral characteristics of ground features. Because the target panchromatic image has only one very wide continuous band (typically covering visible to near-infrared), its low-frequency subband represents an average radiant intensity or overall brightness over a broad spectral range, making it impossible to distinguish the differences in normalized pixel values ​​across different bands. It can only tell us "how bright" a ground feature is overall, but cannot identify which ground feature is causing this brightness. In contrast, the target multispectral image's low-frequency subband is calculated separately across multiple narrow bands (such as blue, green, red, and near-infrared). Therefore, it fully preserves the unique spectral normalized pixel value curves of different ground feature types, which is the core basis for distinguishing vegetation, water bodies, bare soil, and other ground features.

[0066] Based on this, when generating the target fusion image, the low-frequency sub-band of the target multispectral image can be used entirely, while the low-frequency sub-band of the target panchromatic image can be discarded, thus ensuring the spectral authenticity of the target fusion image.

[0067] Optionally, multiple fused high-frequency subbands and low-frequency subbands of the target multispectral image can be subjected to NSCT inverse transformation. By cascading non-downsampling directional synthesis filter banks and non-downsampling tower synthesis filter banks, optimal reconstruction of the target fused image can be achieved at multiple scales and in multiple directions.

[0068] Specifically, for each decomposition scale, the high-frequency subbands obtained from the fusion at that scale are input into a non-downsampling directional synthesis filter bank for frequency domain directional energy aggregation to generate a high-frequency approximation signal for the corresponding scale. Then, using a dual-channel non-decimation tower synthesis filter bank, the high-frequency approximation signal of each scale is recursively merged with the low-frequency component from the previous scale, from the finest scale to the coarsest scale, without any upsampling operation throughout the process. Finally, a high-resolution target fusion image is output.

[0069] It should be understood that because the NSCT inverse transform employs non-downsampling operations throughout, it ensures strict translation invariance during reconstruction, effectively avoiding the Gibbs effect and phase distortion. Simultaneously, through cascaded directional synthesis and tower synthesis filter banks, complete reconstruction of the fused high-frequency coefficients and retained low-frequency coefficients across all scales and directions is achieved, completing the lossless inversion from overcomplete contour wave coefficients to radiometrically consistent pixels. The reconstructed target fused image retains the spectral characteristics of the target multispectral image while incorporating the sharp spatial details of the target panchromatic image.

[0070] S105. Determine the region boundary based on the target fusion image.

[0071] Understandably, a geographic information system (GIS) software platform can use target fused imagery as a high-precision base map, accurately overlay it with legally mandated planning control line vector data, and perform logical analysis. It can also resolve potential spatial conflicts based on preset priority rules, thereby generating a legal, compliant, and conflict-free final boundary delineation map.

[0072] Therefore, the target image can be input into a professional geographic information system software platform (such as ArcGIS or ENVI), and the regional boundary can be determined based on the geographic information system software platform.

[0073] Specifically, first, the target fusion image and various planning control line vector data serving as constraints are input into the software. These vector data include at least the municipal-level urban development boundary, ecological protection red line, permanent basic farmland control line, and nature reserve control line, which can be manually set.

[0074] Optionally, the vector data can be unified to the same coordinate system as the target panchromatic image and the target multispectral image to ensure that all vector data are fully registered with the target fused image in a spatial reference system.

[0075] Next, vector overlay and pixel masking operations are performed. The planned control line vector layer is overlaid on the target fused image, and the fused image is cut based on the boundaries of the vector polygons to generate image blocks corresponding to the boundaries of each control line. Subsequently, critical overlapping portion removal and boundary adjustments are performed.

[0076] In this embodiment, by co-processing the target panchromatic image and the target multispectral image at multiple scales, the spatial detail advantage of the panchromatic image and the spectral information advantage of the multispectral image are fully utilized. By adopting a differentiated fusion strategy for each pair of high-frequency sub-bands based on the gradient difference and boundary saliency difference, features with clearer boundaries and textures can be identified and fused. While improving resolution, the spectral characteristics of ground features are maintained, and the area boundaries delineated based on this are more accurate and reliable.

[0077] Combination Figure 1 ,like Figure 2 As shown, in one implementation of this application embodiment, a high-frequency sub-band includes high-frequency coefficients of multiple pixels, and the pixels in a pair of high-frequency sub-bands correspond one-to-one. The above-mentioned S103 can be specifically implemented by S201-S203.

[0078] S201. Based on the gradient difference between each pair of pixels in each pair of high-frequency subbands, determine the spatial weights of the two pixels included in each pair of pixels.

[0079] It is understandable that the pixels included in each of the two high-frequency subbands in a pair are also in one-to-one correspondence. That is, at a pixel location, its high-frequency coefficient in the target panchromatic image and its high-frequency coefficient in the target multispectral image can be found respectively.

[0080] In this embodiment, when the details represented by the high-frequency coefficient of a pixel are clear in edge and continuous in direction, the higher spatial weight of the high-frequency coefficient of the pixel is used to retain clearer geometric information.

[0081] In one alternative implementation, the gradient magnitudes of the two pixels included in a pair of pixels can be normalized, and the resulting values ​​can be determined as their respective spatial weights.

[0082] S202. Based on the boundary significance difference of each pair of pixels in each pair of high-frequency subbands, determine the spatial injection weights of the two pixels included in each pair of pixels.

[0083] Understandably, when the details represented by the high-frequency coefficients of a pixel can indicate that it is located on the actual boundary of the ground feature, spatial weights can be injected into the pixel with higher high-frequency coefficients to ensure that the injected data is an effective boundary rather than an artifact.

[0084] It should be understood that the boundary saliency difference of a pair of pixels is the difference between the boundary saliency of the two pixels included in that pair of pixels.

[0085] In one alternative implementation, the sum of the boundary saliencies of the two pixels included in a pair of pixels can be determined as the total boundary saliency, and the ratio of the boundary saliency of each pixel in the pair to the total boundary saliency can be determined as the respective spatial injection weight.

[0086] S203. Based on the spatial weights and spatial injection weights of the two pixels in each pair of pixels, the high-frequency coefficients of each pair of pixels are fused to obtain the high-frequency coefficients of each fused pixel in multiple fused high-frequency subbands.

[0087] In one optional implementation, based on the spatial weights of the first panchromatic pixel in the first panchromatic high-frequency sub-band and the first multispectral pixel in the first multispectral high-frequency sub-band, the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel are fused to obtain the spatial fusion coefficient of the first fused pixel in the high-frequency sub-band after the fusion of the first panchromatic high-frequency sub-band and the first multispectral high-frequency sub-band; based on the spatial injection weights of the first panchromatic pixel and the first multispectral pixel, the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel are fused to obtain the spatial injection fusion coefficient of the first fused pixel; the spatial fusion coefficient and the spatial injection fusion coefficient of the first fused pixel are fused to obtain the high-frequency coefficient of the first fused pixel.

[0088] Wherein, the first panchromatic pixel and the first multispectral pixel are a pair of pixels, the first panchromatic pixel is any pixel in the first panchromatic high-frequency sub-band, and the first multispectral pixel is any pixel in the first multispectral high-frequency sub-band.

[0089] Optionally, the spatial fusion coefficient of a fused pixel satisfies the following formula:

[0090]

[0091] in, Indicates merged pixels Spatial fusion coefficient, Represents full-color pixels Spatial weights, Represents full-color pixels High frequency coefficients, Represents multispectral pixels Spatial weights, Represents multispectral pixels The high frequency coefficients.

[0092] In this formula, Represents spatial location coordinates. Since a pair of pixels and its corresponding merged pixel are at the same pixel location, their coordinates are the same. Indicates scale index. Indicates direction index, Represents the target panchromatic image. Indicates the target's multispectral image. Indicates the target panchromatic image corresponding to scale Full-color pixels in the high-frequency sub-band of the direction, Indicates the target multispectral image corresponding to scale Multispectral pixels in the high-frequency subband of the direction.

[0093] Based on the spatial fusion coefficients calculated by this formula, the injection ratio of high-frequency coefficients of the two images is intelligently adjusted according to the spatial quality comparison between the target panchromatic image and the target multispectral image at each pixel location, forming an intermediate fusion result with optimal spatial detail, thus realizing adaptive detail fusion based on spatial quality assessment.

[0094] Optionally, the spatial injection fusion coefficient of a fused pixel satisfies the following formula:

[0095]

[0096] in, Indicates merged pixels The spatial injection fusion coefficient is used to characterize the homogeneity features based on multispectral images, and represents the injection result after conditionally filtering the spatial details of panchromatic images. Represents full-color pixels Spatial injection weights, Represents full-color pixels High frequency coefficients, Represents multispectral pixels Spatial injection weights, Represents multispectral pixels The high frequency coefficients.

[0097] Based on the spatial injection fusion coefficients calculated by this formula, the injection ratio of the high-frequency coefficients of the two images is intelligently adjusted according to the spectral quality comparison between the target panchromatic image and the target multispectral image at each pixel location, forming the optimal intermediate fusion result in terms of spectral detail, thus realizing adaptive detail fusion based on spectral quality assessment.

[0098] Optionally, the spatial injection fusion coefficient and the mean of the spatial fusion coefficients of a fused pixel can be used to determine the high-frequency coefficients of that fused pixel.

[0099] Understandably, while the resolution of the target multispectral image is not high, it possesses value in terms of "fidelity" and "reference." Based on the target multispectral image, the process of injecting details included in the target panchromatic image can be transformed into a controlled and conditional enhancement process, rather than a crude replacement. When the details in the target panchromatic image are reliable (high spatial weights and spatial injection weights), the fusion result closely approximates the sharp details of the target panchromatic image; when the details in the target panchromatic image are unreliable (e.g., in shadow or noisy areas), the fusion result automatically reverts to the original details of the target multispectral image. Although not sharp, this maximizes spectral fidelity and avoids introducing severe distortion.

[0100] In this embodiment, spatial fusion coefficients and spatial injection fusion coefficients are generated separately to avoid mutual interference between the two types of information in the traditional single fusion process. By constructing a dual fusion of spatial fusion coefficients and spatial injection fusion coefficients, the rich texture details in the panchromatic image can be fully explored, and the unique spectral features of the multispectral image can be effectively protected. This achieves the separation optimization and synergistic integration of spatial detail enhancement and spectral feature preservation.

[0101] In one alternative implementation, the spatial fusion coefficient and spatial injection fusion coefficient of the first fused pixel can be fused based on preset dynamic balance parameters to obtain the high-frequency coefficient of the first fused pixel.

[0102] Optionally, the high-frequency coefficients of a fused pixel satisfy the following formula:

[0103]

[0104] in, Indicates merged pixels High frequency coefficients, express The weight, , which are preset dynamic balance parameters, Indicates merged pixels Spatial fusion coefficient, Indicated Weight, Indicates merged pixels Spatial injection fusion coefficient.

[0105] Based on this formula, it should be understood that and These represent candidate fusion feature maps with two different optimization orientations: It is the result of optimization in the dimension of spatial detail, and its goal is to maximize the integration of the sharp edges and textures of the panchromatic image; This is the result optimized in terms of spectral fidelity, with the goal of preserving the spectral continuity and ground cover separability of multispectral imagery to the maximum extent. By constructing two independent feature channels, spatial detail enhancement and spectral feature preservation are separated and optimized.

[0106] Therefore, the resulting fused pixels are not one of the direct outputs, but rather based on preset dynamic balancing parameters. The two candidate feature maps are then weighted and fused at the feature level to obtain the final high-frequency coefficients. This process is essentially a decision-making or trade-off process. It adaptively selects and fuses optimization results from two different dimensions based on regional characteristics (e.g., peri-urban areas require greater emphasis on spectral fidelity to suppress artifacts, while ecological reserves require greater emphasis on spatial detail to capture natural boundaries), rather than directly weighting the original panchromatic and multispectral high-frequency coefficients. This method gives the fusion process a clearer physical meaning and better controllability, enabling flexible responses to the needs of different terrain scenarios.

[0107] In this embodiment, the value of the preset dynamic balance parameter is determined by the region type where the fused pixel is located.

[0108] Optionally, land use data of the area where the fused pixel is located can be obtained, and the area can be divided into urban-rural fringe areas, built-up land, regional ecological protection zones, and farmland areas based on its functional attributes. In urban-rural fringe areas and built-up land, the shadows and roof structures have a greater impact on the edges, therefore... The value is set to 0.4, allowing the spatial injection blending coefficient to play a major role in removing false edges such as shadows and rooftops. In ecological protection areas and farmland areas, the boundaries are mostly natural boundaries with fewer false edges, and the spectral transition of the true boundaries may be relatively gentle, so the value can be set to 0.4. The value is set to 0.6, allowing the spatial weights to play a major role in ensuring that all subtle real boundaries are captured.

[0109] Optionally, the value of the preset dynamic balance parameter can also be modified according to actual needs.

[0110] Understandably, preset dynamic balance parameters can provide an important control dimension for the fusion process, enabling flexible adjustment of the priority between spatial detail and spectral fidelity according to the specific needs of different application scenarios.

[0111] In another alternative implementation, the average of the spatial weight and the spatial injection weight can be determined as the fusion weight. Then, the high-frequency coefficients are weighted and fused based on the fusion weight of each pixel in a pair of pixels to obtain the high-frequency coefficients of each fused pixel.

[0112] The methods provided in S201-S203 calculate spatial weights and spatial injection weights for each pixel, respectively, and adjust the fusion strategy according to the specific image features of each pixel. This refines the fusion granularity from the traditional sub-band level to the pixel level, greatly improving the precision and adaptability of the fusion process. This allows for clearer preservation of the boundaries of small features such as road edges and field ridges, while effectively suppressing noise amplification in uniform feature areas. By implementing pixel-level weight calculation and fusion mechanisms, the method avoids the loss of detail or spectral distortion caused by a one-size-fits-all fusion approach.

[0113] Combination Figure 2 ,like Figure 3 As shown, in one implementation of this application embodiment, the above-mentioned S201 can be specifically implemented by S301-S304.

[0114] S301. Determine the magnitude weight of the first panchromatic pixel based on the ratio of the gradient magnitude of the local region where the first panchromatic pixel is located to the total gradient magnitude.

[0115] The total gradient magnitude is the sum of the gradient magnitude of the local region where the first multispectral pixel is located and the gradient magnitude of the local region where the first panchromatic pixel is located.

[0116] Optionally, the local area can be a 5×5 area centered on the first full-color pixel.

[0117] Optionally, the magnitude weight of a full-color pixel satisfies the following formula:

[0118]

[0119] in, Represents full-color pixels Amplitude weights, Represents full-color pixels The gradient magnitude of the local region where it is located, Represents multispectral pixels The gradient magnitude of the local region where it is located, This represents the total gradient magnitude.

[0120] Optionally, when the denominator in the above formula is 0, it indicates that the gradient magnitude at that pixel location is 0, and it is impossible to evaluate whether the gradient feature is significant. In this case, the gradient magnitude can be... The value is set to 0, meaning that the subsequent process relies entirely on spatial injection weights for fusion.

[0121] It should be understood that the ratio of the gradient magnitude of the local region where the first panchromatic pixel is located to the total gradient magnitude represents the significance of the gradient feature of the first panchromatic pixel within that local region. The larger the gradient magnitude of the local region where the first panchromatic pixel is located, the greater the magnitude weight of the first panchromatic pixel.

[0122] In one alternative implementation, the gradient values ​​of each pixel in the horizontal and vertical directions can be calculated using the Sobel operator. Then, the gradient magnitude of each pixel is determined based on these horizontal and vertical gradient values. Finally, the gradient magnitude of the region is determined based on the gradient magnitudes of all pixels within the region containing each pixel.

[0123] Optionally, the gradient magnitude of a pixel satisfies the following formula:

[0124]

[0125] in, Represents pixels gradient magnitude, Represents pixels The gradient value in the horizontal direction, Represents pixels The gradient value in the vertical direction.

[0126] S302. Determine the orientation consistency weight of the first panchromatic pixel based on the standard deviation of the gradient orientation angle of the first panchromatic pixel.

[0127] In one alternative implementation, the gradient direction angle of each pixel can be obtained, and then the standard deviation of the gradient direction angle of the first panchromatic pixel can be determined based on the standard deviation of the gradient direction angles of all pixels in the local region.

[0128] It should be understood that the smaller the standard deviation of the gradient direction angle of the first full-color pixel, the more consistent the gradient direction is in that local area, and the more likely that the local area is a continuous and real boundary; the larger the standard deviation, the more chaotic the gradient direction is in that local area, and the more likely it is a noisy or complex texture area.

[0129] Optionally, the orientation consistency weight of a pixel satisfies the following formula:

[0130]

[0131] in, Represents full-color pixels directional consistency weight, Represents full-color pixels The standard deviation of the gradient direction angle, Represents full-color pixels The variance of the gradient direction angle, This is an empirical threshold used to regulate the sensitivity to gradient direction consistency.

[0132] The value can be set according to the expected shape characteristics of the target ground feature boundary.

[0133] For example, when the boundaries of features in a local area are mainly regular and straight (such as roads, field ridges, and building outlines). A lower value (11.5 to 22.9 degrees) is chosen to make the gradient direction highly sensitive to changes, assigning high weight only to edges with extremely consistent directions. This rigorously selects the most regular linear boundaries and effectively suppresses curved natural edges or noise. When the boundaries of local features are dominated by meandering, irregular natural contours (such as forest edges or lake shorelines), a higher value (34.3 to 45.8 degrees) is chosen to make the gradient direction more tolerant of changes, allowing for some natural curvature of the boundaries. This ensures that the continuous natural contours are captured and preserved intact, avoiding any fragmentation.

[0134] Based on this formula, it should be understood that when When it is close to 1, it indicates that the directions are extremely consistent; when When the value is close to 0, it indicates that the directions are extremely inconsistent.

[0135] S303. The product of the directional consistency weight and the amplitude weight is determined as the spatial weight of the first panchromatic pixel.

[0136] Optionally, the spatial weight of a full-color pixel satisfies the following formula:

[0137]

[0138] in, Represents full-color pixels Spatial weights, Represents full-color pixels directional consistency weight, Represents full-color pixels The magnitude weight.

[0139] In this formula, the physical rationality of the edge is verified by the orientation consistency weight, the random high-frequency response caused by noise is suppressed, and the intensity contrast of the edge is evaluated by the amplitude weight, so that significant spatial details are preferentially injected. This ensures that only true boundaries with continuous orientation and significant amplitude are given high weight, and that the highest quality and most reliable spatial details in the panchromatic image are preferentially and fully injected into the fusion result, which greatly improves the accuracy of boundary recognition.

[0140] S304. Determine the spatial weight of the first multispectral pixel based on the spatial weight of the first panchromatic pixel.

[0141] The sum of the spatial weight of the first panchromatic pixel and the spatial weight of the first multispectral pixel is 1.

[0142] It should be understood that the larger the spatial weight of the first panchromatic pixel, the smaller the spatial weight of the first multispectral pixel. Since the gradient information of the panchromatic pixel is more significant and accurate, it is possible to improve the computational efficiency by only calculating the spatial weight of the panchromatic pixel and then comparing it with the spatial weight of the first panchromatic pixel to obtain the spatial weight of the first multispectral pixel.

[0143] The methods provided in S301-S304 above first preliminarily assess edge saliency by calculating the gradient magnitude of each pair of pixels. Then, gradient direction consistency is introduced as a key constraint. By measuring the degree of coordination of gradient directions within the pixel neighborhood, high-quality edge points belonging to continuous, real-world object boundaries are effectively identified, while suppressing isolated high-frequency responses caused by noise or cluttered textures. Finally, gradient magnitude and direction consistency are integrated into the weight calculation, ensuring that edge regions with both high saliency and high continuity in the panchromatic image are assigned the highest weight. This injects sharp geometric details while maximizing the integrity and smoothness of the boundaries, providing the optimal geometric foundation for subsequent accurate boundary recognition.

[0144] Understandably, based on the boundary land cover types, land cover can be broadly categorized into straw, roads, water bodies, and vegetation, with straw existing in a transitional state between vegetation (green) and bare soil (brown). In the visible light band, the normalized pixel value is higher than that of healthy vegetation, while in the near-infrared band, the normalized pixel value is significantly lower than that of healthy vegetation. The spectral curve of roads is flat, without obvious absorption or reflection peaks. The normalized pixel value is generally moderately high, and does not decrease sharply in the near-infrared and short-wave infrared bands. Water bodies exhibit strong absorption in the near-infrared and short-wave infrared bands, resulting in extremely low normalized pixel values; their normalized pixel values ​​are relatively high in the visible blue-green band. Vegetation exhibits strong absorption (low reflectance) in the red light band and strong reflection in the near-infrared band, forming a steep "red edge" effect.

[0145] Traditional spatial injection weighting analysis relies solely on the spectral values ​​of individual pixels, failing to adequately consider the spatial continuity and internal heterogeneity of ground features. To address this issue, this application introduces image segmentation and spatial autocorrelation analysis to more accurately assess the discriminative power of panchromatic and multispectral images at real ground feature boundaries, thereby calculating superior spatial injection weights.

[0146] Combination Figure 2 ,like Figure 4 As shown, in one implementation of this application embodiment, the above-mentioned S202 can be specifically implemented by S401-S405.

[0147] S401. Perform image segmentation on the target panchromatic image and the target multispectral image to obtain multiple homogeneous regions included in each image.

[0148] Specifically, an image is divided into multiple homogeneous regions using multiple dividing lines.

[0149] Optionally, the target panchromatic image and the target multispectral image can be co-segmented based on a multi-resolution segmentation algorithm to obtain multiple homogeneous regions with clear boundaries within each image. Each homogeneous region represents a potential ground feature unit. It should be noted that this multi-resolution segmentation algorithm uses the spectral and spatial features of the images as criteria, aggregating adjacent pixels with similar spectral and spatial features into homogeneous image objects, i.e., homogeneous regions. For the target panchromatic image, segmentation is mainly based on its grayscale texture features; for the target multispectral image, segmentation is based on its multi-band spectral features.

[0150] S402. Determine the boundary saliency weight of the first panchromatic pixel based on the ratio of the boundary saliency of the first panchromatic homogeneous region where the first panchromatic pixel is located to the total boundary saliency.

[0151] The total boundary saliency is the sum of the boundary saliency of the first panchromatic homogeneous region and the boundary saliency of the first multispectral homogeneous region.

[0152] It should be understood that the first multispectral homogeneous region is the homogeneous region where the first multispectral pixel is located.

[0153] In this embodiment of the application, the boundary salience of a homogeneous region containing a pixel is determined as the boundary salience of that pixel.

[0154] In one optional implementation, the normalized pixel value of each homogeneous region included in the target panchromatic image is determined; the absolute value of the difference between the normalized pixel value within the first panchromatic homogeneous region and the mean of the normalized pixel values ​​of its neighboring homogeneous regions is determined as the boundary salience of the first panchromatic homogeneous region. It should be understood that neighboring homogeneous regions of a homogeneous region are other homogeneous regions adjoining the boundary of that homogeneous region, and the number of neighboring homogeneous regions of a homogeneous region depends on the number of homogeneous regions adjoining the boundary of that homogeneous region.

[0155] Similarly, the boundary saliency of the first multispectral homogeneous region is calculated. Since the first multispectral homogeneous region has multiple bands, the characteristic bands can be determined based on the land cover type of the homogeneous region, and the boundary saliency of the first multispectral homogeneous region can be determined based on the normalized pixel values ​​of the characteristic bands.

[0156] For example, vegetation has the highest normalized pixel value in the near-infrared band, forming the strongest contrast with most non-vegetated features (such as soil and water bodies), and best highlighting the boundaries of vegetated areas. The difference in normalized pixel values ​​of vegetation in the near-infrared band is usually large; therefore, the normalized pixel values ​​of vegetation in the near-infrared band can be used to determine the salience of the boundary. Water bodies have normalized pixel values ​​of almost zero in the near-infrared and short-wave infrared bands, showing a great difference from the surrounding features; the normalized pixel values ​​of the near-infrared or short-wave infrared bands can be used to determine the salience of the boundary. The green light band helps to distinguish bare soil from vegetation (vegetation absorbs red light but reflects green light), and the near-infrared band helps to distinguish bare soil from shadows or water bodies in terms of overall brightness; therefore, the normalized pixel values ​​of the green light band or near-infrared band can be used to determine the salience of the boundary. Straw has a unique response in the short-wave infrared that is different from healthy vegetation and bare soil; the normalized pixel values ​​of the short-wave infrared band can be used to determine the salience of the boundary.

[0157] Optionally, the boundary saliency weight of a full-color pixel satisfies the following formula:

[0158]

[0159] in, Represents full-color pixels Boundary significance weights, Represents full-color pixels Boundary saliency, Represents multispectral pixels Boundary saliency, This indicates the significance of the overall boundary.

[0160] Based on this formula, it should be understood that the higher the boundary saliency of a panchromatic pixel, the greater the boundary saliency weight of that panchromatic pixel.

[0161] Optionally, when the denominator in the above formula is 0, it can be determined that... The value is 0, meaning that fusion relies entirely on spatial weights. The value and When all values ​​are 0, the details of the target panchromatic image are not injected into the target multispectral image, and the target multispectral image is determined as the target fused image.

[0162] It is understandable that the essence of the boundary of a feature is the significant difference in the attributes of adjacent areas. Therefore, by calculating the relative ratio of the difference in the mean of normalized pixel values ​​between areas, the significance of the boundary at different locations can be quantified more accurately.

[0163] S403. Based on the spectral variance and Moran index in the first multispectral homogeneous region where the first multispectral pixel is located, determine the dynamic weighting factor of the first multispectral pixel.

[0164] It should be understood that the spectral variance within a homogeneous region is used to measure the dispersion of the spectral values ​​of pixels within that region. For example, a pure forest has a small spectral variance, while a farmland area containing vegetation of varying health conditions or mixed soil has a large variance.

[0165] Optionally, the variance of the normalized pixel values ​​of all pixels in a homogeneous region on the same band can be calculated first, and then the mean of the variances of all bands can be determined as the spectral variance.

[0166] Understandably, the Moran index is used to measure the spatial autocorrelation of pixel spectral values ​​within a homogeneous region. The higher the Moran index, the more homogeneous the region.

[0167] For example, continuous land features (such as large bodies of water, lush vegetation, and homogeneous roads) have spatially clustered pixel values, and the Moran index is significantly positive, close to 1. For highly heterogeneous or fragmented areas, the Moran index tends to be close to 0 or negative.

[0168] Optionally, since the Moran index may be negative, the Moran index of multiple homogeneous regions in the target multispectral image can be subjected to max-min normalization to map its value to [0, 1].

[0169] In one alternative implementation, the homogeneity confidence of the first multispectral homogeneous region can be determined based on the spectral variance and Moran's index of the first multispectral homogeneous region where the first multispectral pixel is located; the ratio of the homogeneity confidence of the first multispectral homogeneous region to the maximum homogeneity confidence of the homogeneous regions included in the multispectral image is determined as the dynamic weighting factor of the first multispectral pixel.

[0170] It should be understood that the homogeneity confidence score of a multispectral homogeneous region is used to characterize the reliability or probability that the region contains only the same type of land cover. When the homogeneity confidence score is high, it indicates that the region is a reliable, single land cover unit. Therefore, the boundary saliency calculated at the boundary of this region is highly reliable and likely represents the actual land cover variation. Consequently, the system assigns a higher spatial injection weight, thus injecting more sharp details from the panchromatic image during fusion. Conversely, when the homogeneity confidence score is low, it indicates that the region itself is highly heterogeneous. In this case, the calculated boundary differences may simply be internal noise rather than the true boundary. In this situation, the contribution of the spatial injection weight can be reduced, relying more on the spatial weight to avoid introducing erroneous details.

[0171] Optionally, the homogeneity confidence of a homogeneous region satisfies the following formula:

[0172]

[0173] in, Indicates homogeneous regions Homogeneity confidence level Indicates homogeneous regions Moran's index, Indicates homogeneous regions Spectral variance, This represents the maximum spectral variance of all homogeneous regions in the target multispectral image. This represents an exponential function.

[0174] In this formula, The larger, The smaller, The smaller. Used for Normalize.

[0175] Optionally, the ratio between the homogeneity confidence of the first multispectral homogeneous region and the maximum value of all multispectral homogeneous regions in the target multispectral image can be determined as the dynamic weighting factor of the first multispectral homogeneous region. This ensures that when the homogeneity of the first multispectral homogeneous region is high, the dynamic weighting factor approaches 1, and the spatial injection weight plays a greater role; when the homogeneity of the first multispectral homogeneous region is low, the dynamic weighting factor approaches 0, and the contribution of the spatial injection weight is lower, thus avoiding assigning excessively high weights to fragmented regions.

[0176] It should be understood that since panchromatic images do not have spectral features, the homogeneity of the first multispectral homogeneous region is evaluated based on the homogeneity confidence of the first multispectral homogeneous region. Since the first multispectral homogeneous region and the first panchromatic homogeneous region are corresponding, it can represent the homogeneity level of the first panchromatic homogeneous region.

[0177] S404. The product of the dynamic weight factor and the boundary saliency weight is determined as the spatial injection weight of the first full-color pixel.

[0178] It should be understood that this spatial injection weight is used to control the degree of injection of high-frequency spatial details in panchromatic images.

[0179] Understandably, the dynamic weighting factor is used to optimize the boundary saliency weights, so that the spatial injection weights more fully consider the spatial distribution characteristics of land features, and can effectively avoid heterogeneous...

[0180] Fragmented regions are given excessively high spatial injection weights, thereby achieving higher positioning accuracy at real, continuous ground feature boundaries.

[0181] It should be understood that this spatial injection weight is not used to control the preservation of spectral information, but rather to determine whether to inject spatial texture from the panchromatic image based on the homogeneity characteristics of the multispectral image (such as spectral variance and Moran's index). When the homogeneity confidence of a multispectral homogeneous region is high, it indicates that the terrain features within the region are simple and the boundaries are realistic and reliable. In this case, a higher spatial injection weight should be assigned to fully inject the sharp spatial details of the panchromatic image. Conversely, when the homogeneity confidence is low, it indicates that the region is mixed and the boundaries may be noise. In this case, the spatial injection weight should be reduced to avoid introducing erroneous spatial details.

[0182] S405. Determine the spatial injection weight of the first multispectral pixel based on the spatial injection weight of the first panchromatic pixel.

[0183] The sum of the spatial injection weight of the first panchromatic pixel and the spatial injection weight of the first multispectral pixel is 1.

[0184] It should be understood that the larger the spatial injection weight of the first panchromatic pixel, the smaller the spatial injection weight of the first multispectral pixel, and the sum of the two is always equal to 1.

[0185] Based on the methods described in S401-S405, by combining the spectral statistical characteristics and spatial autocorrelation properties of homogeneous regions, the reliability of each homogeneous region representing a single land feature can be accurately assessed, and the confidence level of the boundary saliency weight can be dynamically adjusted accordingly. This can effectively avoid the problem of incorrectly assigning high spatial injection weights in heterogeneous regions and significantly improve the spectral continuity preservation effect at land feature boundaries.

[0186] Please see Figure 5 The diagram illustrates the structure of a land spatial planning area boundary determination system based on remote sensing imagery, provided in one embodiment of this application.

[0187] like Figure 5 As shown, the land spatial planning area boundary determination system 50 based on remote sensing imagery includes an image acquisition module 501, an image processing module 502, and a boundary determination module 503.

[0188] Image acquisition module 501 is used to acquire spatially aligned multispectral and panchromatic images of the target region.

[0189] The image processing module 502 is used to perform multi-scale and multi-directional transformations on the target panchromatic image and the target multispectral image to obtain the low-frequency sub-band and multiple high-frequency sub-bands corresponding to each image.

[0190] Among them, multiple high-frequency sub-bands on the target panchromatic image correspond one-to-one with multiple high-frequency sub-bands on the target multispectral image.

[0191] The image processing module 502 is also used to fuse each pair of high-frequency subbands based on the gradient difference and boundary saliency difference between each pair of high-frequency subbands to obtain multiple fused high-frequency subbands.

[0192] The image processing module 502 is also used to perform multi-scale and multi-directional inverse transformation based on multiple fused high-frequency subbands and low-frequency subbands of the target multispectral image to obtain the target fused image.

[0193] Boundary determination module 503 is used to determine the region boundary based on the target fused image.

[0194] Optionally, the remote sensing image-based land spatial planning area boundary determination system 50 can be used to perform any of the above-mentioned remote sensing image-based land spatial planning area boundary determination methods.

[0195] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0196] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for determining the boundary of a land spatial planning area based on remote sensing imagery, characterized in that, include: Acquire spatially aligned multispectral and panchromatic images of the target region; Multi-scale and multi-directional transformations are performed on the target panchromatic image and the target multispectral image to obtain the low-frequency sub-band and multiple high-frequency sub-bands corresponding to each image. The multiple high-frequency sub-bands on the target panchromatic image correspond one-to-one with the multiple high-frequency sub-bands on the target multispectral image. A high-frequency sub-band includes the high-frequency coefficients of multiple pixels, and the pixels in a pair of high-frequency sub-bands correspond one-to-one. Based on the gradient difference and boundary significance difference between each pair of high-frequency subbands, each pair of high-frequency subbands is fused to obtain multiple fused high-frequency subbands, including: Based on the gradient difference between each pair of pixels in each pair of high-frequency subbands, the spatial weights of the two pixels in each pair are determined, including: determining the magnitude weight of the first panchromatic pixel based on the ratio of the gradient magnitude of the local region where the first panchromatic pixel is located to the total gradient magnitude, wherein the total gradient magnitude is the sum of the gradient magnitudes of the local region where the first multispectral pixel is located and the gradient magnitudes of the local region where the first panchromatic pixel is located; determining the directional consistency weight of the first panchromatic pixel based on the standard deviation of the gradient direction angle of the first panchromatic pixel; determining the spatial weight of the first panchromatic pixel by multiplying the directional consistency weight and the magnitude weight; determining the spatial weight of the first multispectral pixel based on the spatial weight of the first panchromatic pixel; and the sum of the spatial weights of the first panchromatic pixel and the first multispectral pixel is 1. Based on the boundary significance difference of each pair of pixels in each pair of high-frequency subbands, the spatial injection weights of the two pixels included in each pair of pixels are determined. Based on the spatial weights and spatial injection weights of the two pixels in each pair, the high-frequency coefficients of each pair of pixels are fused to obtain the high-frequency coefficients of each fused pixel in multiple fused high-frequency sub-bands. This includes: fusing the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel in the first panchromatic high-frequency sub-band based on the spatial weights of the first panchromatic pixel and the first multispectral pixel in the first multispectral high-frequency sub-band to obtain the spatial fusion coefficient of the first fused pixel in the high-frequency sub-band after the fusion of the first panchromatic high-frequency sub-band and the first multispectral high-frequency sub-band. The first panchromatic pixel and the first multispectral pixel are a pair of pixels. Based on the spatial injection weights of the first panchromatic pixel and the first multispectral pixel, the high-frequency coefficients of the first panchromatic pixel and the first multispectral pixel are fused to obtain the spatial injection fusion coefficient of the first fused pixel. The spatial fusion coefficient and the spatial injection fusion coefficient of the first fused pixel are then fused to obtain the high-frequency coefficient of the first fused pixel. Multi-scale, multi-directional inverse transform is performed based on multiple fused high-frequency subbands and low-frequency subbands of the target multispectral image to obtain the target fused image; The region boundary is determined based on the target fused image.

2. The method for determining the boundary of a land spatial planning area based on remote sensing imagery according to claim 1, characterized in that, The spatial fusion coefficient and spatial injection fusion coefficient of the first fused pixel are fused to obtain the high-frequency coefficients of the first fused pixel, including: The spatial fusion coefficient and spatial injection fusion coefficient of the first fused pixel are fused based on preset dynamic balance parameters to obtain the high-frequency coefficient of the first fused pixel.

3. The method for determining the boundary of a land spatial planning area based on remote sensing imagery according to claim 1, characterized in that, Based on the boundary saliency difference of each pair of pixels in each pair of high-frequency subbands, the spatial injection weights of the two pixels included in each pair of pixels are determined, including: Image segmentation is performed on the target panchromatic image and the target multispectral image to obtain multiple homogeneous regions included in each image; The boundary saliency weight of the first panchromatic homogeneous region is determined based on the ratio of the boundary saliency of the first panchromatic homogeneous region where the first panchromatic pixel is located to the total boundary saliency, wherein the total boundary saliency is the sum of the boundary saliency of the first panchromatic homogeneous region and the boundary saliency of the first multispectral homogeneous region. The dynamic weighting factor of the first multispectral pixel is determined based on the spectral variance and Moran index in the first multispectral homogeneous region where the first multispectral pixel is located. The product of the dynamic weight factor and the boundary saliency weight is determined as the spatial injection weight of the first panchromatic pixel. The spatial injection weight of the first multispectral pixel is determined based on the spatial injection weight of the first panchromatic pixel, and the sum of the spatial injection weight of the first panchromatic pixel and the spatial injection weight of the first multispectral pixel is 1.

4. The method for determining the boundary of a land spatial planning area based on remote sensing imagery according to claim 1 or 3, characterized in that, The method further includes: Determine the normalized pixel value for each homogeneous region included in the target panchromatic image; The absolute value of the difference between the normalized pixel value in the first panchromatic homogeneous region and the mean of the normalized pixel values ​​in the adjacent homogeneous regions of the first panchromatic homogeneous region is determined as the boundary salience of the first panchromatic homogeneous region.

5. The method for determining the boundary of a land spatial planning area based on remote sensing imagery according to claim 3, characterized in that, The determination of the dynamic weighting factor for the first multispectral pixel based on the spectral variance and Moran's index in the first multispectral homogeneous region where the first multispectral pixel is located includes: Based on the spectral variance and Moran's index of the first multispectral homogeneous region where the first multispectral pixel is located, the homogeneity confidence of the first multispectral homogeneous region is determined. The ratio of the homogeneity confidence of the first multispectral homogeneous region to the maximum homogeneity confidence of the homogeneous regions included in the multispectral image is determined as the dynamic weighting factor of the first multispectral pixel.

6. The method for determining the boundary of a land spatial planning area based on remote sensing imagery according to claim 1, characterized in that, The method further includes: Acquire the raw multispectral and raw panchromatic images of the target area; The original multispectral image and the original panchromatic image are converted to the same coordinate system; The original multispectral image is resampled based on the resolution of the original panchromatic image to obtain an intermediate multispectral image with the same size as the original panchromatic image. The original panchromatic image and the intermediate multispectral image are aligned to obtain the target panchromatic image and the target multispectral image.

7. A system for determining the boundary of a land spatial planning area based on remote sensing imagery, characterized in that, The system is used to execute the method for determining the boundary of a land spatial planning area based on remote sensing imagery as described in any one of claims 1-6. The system includes an image acquisition module, an image processing module, and a boundary determination module. The image acquisition module is used to acquire spatially aligned multispectral images and panchromatic images of the target region. The image processing module is used to perform multi-scale and multi-directional transformations on the target panchromatic image and the target multispectral image to obtain a low-frequency sub-band and multiple high-frequency sub-bands corresponding to each image. The multiple high-frequency sub-bands on the target panchromatic image correspond one-to-one with the multiple high-frequency sub-bands on the target multispectral image. The image processing module is also used to fuse each pair of high-frequency subbands based on the gradient difference and boundary saliency difference between each pair of high-frequency subbands to obtain multiple fused high-frequency subbands. The image processing module is also used to perform multi-scale and multi-directional inverse transformation based on multiple fused high-frequency sub-bands and the low-frequency sub-band of the target multispectral image to obtain the target fused image. The boundary determination module is used to determine the region boundary based on the target fused image.

Images