Airport abnormal area identification method and device, electronic equipment and readable medium

Abstract

Embodiments of the present disclosure disclose an airport abnormal area identification method and device, electronic equipment and readable medium. A specific embodiment of the method comprises: selecting reference area data and contrast area data; performing registration correction on the reference area remote sensing image and the contrast area remote sensing image to generate a corrected reference area image and a corrected contrast area image; performing area shadow detection on the corrected reference area image and the corrected contrast area image; performing constraint feature fusion on the corrected reference area image and the corrected contrast area image respectively to generate reference area main features and contrast area main features; performing abnormal area identification on the reference area main features and the contrast area main features; and performing verification grading on the candidate abnormal area identification information to obtain airport target area abnormal information. The embodiment can perform shadow detection on the airport area to identify abnormal shadow areas, and perform verification on the abnormal areas through a grid elevation map, thereby improving the safety of the airport area clearance.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of computer technology, and more specifically to methods, apparatus, electronic devices, and readable media for identifying abnormal areas in airports. Background Technology

[0002] Airport areas contain numerous features and facilities directly related to flight safety, such as runways, taxiways, terminals, hangars, and surrounding buildings. These facilities and buildings may change in height and shape due to new construction, temporary installations, changes in building height, or changes in terrain, posing potential threats to airport airspace safety. Currently, anomaly identification in airport areas typically relies on manual inspections, periodic measurements, or shadow changes based on optical remote sensing imagery. However, remote sensing imagery is susceptible to changes in lighting and shadow projection, leading to shadow boundary changes being misidentified as actual feature changes. Airport buildings are densely packed with varying heights, and shadows change significantly with lighting conditions, easily creating brightness differences in the imagery that resemble actual feature changes, resulting in substantial errors in anomaly identification. Furthermore, relying solely on two-dimensional image changes makes it difficult to determine whether an anomaly truly causes height exceeding limits, thus reducing airport airspace safety. Summary of the Invention

[0003] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0004] Some embodiments of this disclosure provide methods, apparatuses, electronic devices, and computer-readable media for identifying abnormal areas in airports to address the technical problems mentioned in the background section above.

[0005] In a first aspect, some embodiments of this disclosure provide a method for identifying abnormal areas in airports. The method includes: selecting airport area data that meets spatiotemporal conditions from an airport area dataset as reference area data and comparison area data, respectively, based on airport target area information and a target time window, wherein each airport area data includes a regional remote sensing image and a regional raster elevation map; registering and correcting the reference area remote sensing image and the comparison area remote sensing image to generate a corrected reference area image and a corrected comparison area image, wherein the reference area remote sensing image is a regional remote sensing image included in the reference area data, and the comparison area remote sensing image is a regional remote sensing image included in the comparison area data; and registering and correcting the reference area remote sensing image and the comparison area remote sensing image... Regional shadow detection is performed on the reference area image and the aforementioned corrected comparison area image to generate shadow information for each region. Each region shadow information includes a region shadow probability map and shadow constraint information. Based on the corresponding region shadow probability maps, constraint feature fusion is performed on the aforementioned corrected reference area image and the aforementioned corrected comparison area image to generate reference region main features and comparison region main features. Anomaly region identification is performed on the aforementioned reference region main features and comparison region main features to generate candidate anomaly region identification information. Based on the comparison area raster elevation map, the reference area raster elevation map, and each shadow constraint information, the candidate anomaly region identification information is verified and graded to obtain airport target area anomaly information.

[0006] Secondly, some embodiments of this disclosure provide an airport anomaly area identification device. The device includes: a selection unit configured to select airport area data that meets spatiotemporal conditions from an airport area dataset as reference area data and comparison area data, respectively, based on airport target area information and a target time window, wherein each airport area data includes regional remote sensing imagery and regional raster elevation map; a registration and correction unit configured to perform registration and correction on the reference area remote sensing imagery and the comparison area remote sensing imagery to generate a corrected reference area imagery and a corrected comparison area imagery, wherein the reference area remote sensing imagery is a regional remote sensing imagery included in the reference area data, and the comparison area remote sensing imagery is a regional remote sensing imagery included in the comparison area data; and an area shadow detection unit configured to detect shadows in the corrected reference area remote sensing imagery. The image and the aforementioned corrected comparison area image are subjected to region shadow detection to generate shadow information for each region. Each region shadow information includes a region shadow probability map and shadow constraint information. A constraint feature fusion unit is configured to perform constraint feature fusion on the aforementioned corrected reference area image and the aforementioned corrected comparison area image respectively based on the corresponding region shadow probability maps to generate reference area main features and comparison area main features. An anomaly region identification unit is configured to identify anomalies on the aforementioned reference area main features and comparison area main features to generate candidate anomaly region identification information. A verification and grading unit is configured to verify and grade the aforementioned candidate anomaly region identification information based on the comparison area raster elevation map, the reference area raster elevation map, and the shadow constraint information to obtain airport target area anomaly information.

[0007] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0008] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0009] The above-described embodiments of this disclosure have the following beneficial effects: the airport anomaly area identification method of some embodiments of this disclosure can perform shadow detection on airport areas to identify abnormal shadow areas, and verify the abnormal areas through raster elevation maps, thereby improving the airspace safety of airport areas. Specifically, the reasons for the reduction in the airspace safety of related airport areas are: remote sensing images are easily affected by factors such as changes in illumination and shadow projection, causing changes in shadow boundaries to be misidentified as changes in real ground features; airport buildings are dense and have large height differences, and shadows change significantly with illumination conditions, easily forming brightness differences in images similar to changes in real ground features, thus leading to large errors in anomaly area identification. At the same time, it is difficult to determine whether anomalies actually cause height exceedances based solely on changes in two-dimensional images, thereby reducing the airspace safety of airport areas. Based on this, the airport anomaly area identification method of some embodiments of this disclosure firstly selects airport area data that meets the spatiotemporal conditions from the airport area dataset as reference area data and comparison area data, respectively, according to airport target area information and target time window. Each airport area data includes regional remote sensing images and regional raster elevation maps. Therefore, based on airport target area information and target time windows, regional data that meets spatiotemporal conditions is selected to ensure consistency in spatial range and temporal scale between the remote sensing images and elevation maps used for comparative analysis, thereby avoiding invalid changes introduced due to inconsistent coverage or excessive time differences. Then, the remote sensing images of the reference area and the comparison area are registered and corrected to generate corrected reference area images and corrected comparison area images. The reference area remote sensing image is the regional remote sensing image included in the reference area data. The comparison area remote sensing image is the regional remote sensing image included in the comparison area data. Thus, by registering and correcting the reference area and comparison area remote sensing images, the spatial position error of the same area in images from different time phases is significantly reduced, thereby reducing geometric displacement caused by imaging angle, orbital deviation, or sensor differences. Subsequently, regional shadow detection is performed on the corrected reference area image and the corrected comparison area image to generate shadow information for each region, where each region shadow information includes a region shadow probability map and shadow constraint information. Therefore, by performing regional shadow detection on the corrected remote sensing image, the distribution of shadows within the airport area caused by differences in building height and changes in lighting conditions can be identified, and corresponding shadow probability maps and shadow constraint information can be generated. This allows for the explicit identification of unstable areas in the image caused by changes in lighting, thereby distinguishing shadow changes from changes in actual ground features. Secondly, based on the corresponding shadow probability maps for each region, constraint feature fusion is performed on the corrected reference region image and the corrected comparison region image to generate the main features of the reference region and the main features of the comparison region.Therefore, under the influence of shadow constraint information, feature fusion can be performed on the corrected reference area image and the corrected contrast area image. This allows the fused image features to suppress changes within shadowed areas and remain sensitive to structural changes within non-shadowed areas. This introduces shadow information into the feature level, rather than simply removing shadowed areas at the result level, improving the anomaly identification process's ability to recognize real structural changes. Next, anomaly region identification is performed on the aforementioned main features of the reference area and the contrast area to generate candidate anomaly region identification information. Thus, by determining the cumulative feature differences between the main features of the reference area and the contrast area after constraint fusion, areas with spatial continuity and significant changes can be identified, forming candidate anomaly region identification information. Finally, based on the aforementioned contrast area raster elevation map, reference area raster elevation map, and various shadow constraint information, the candidate anomaly region identification information is verified and graded to obtain airport target area anomaly information. Therefore, by combining airport target area information, multi-temporal regional raster elevation maps, and shadow constraint information, the height of candidate abnormal areas can be verified to determine whether there is an excessive height or a trend of height change. Based on this, the abnormal areas can be classified into risk levels, thereby improving the stability of abnormal area identification and thus improving the airspace safety within the airport area. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0011] Figure 1 This is a flowchart of some embodiments of the airport anomaly area identification method according to the present disclosure;

[0012] Figure 2 This is a schematic diagram of airport area division in the airport anomaly area identification method disclosed herein;

[0013] Figure 3 This is a schematic diagram of the network structure of the region shadow detection model in the airport anomaly region identification method disclosed herein;

[0014] Figure 4 This is a schematic diagram of the structure of some embodiments of the airport abnormal area identification device according to the present disclosure;

[0015] Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0016] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0017] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0018] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0019] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0020] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0021] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] Figure 1 A flow 100 of some embodiments of an airport anomaly area identification method according to the present disclosure is shown. The airport anomaly area identification method includes the following steps:

[0023] Step 101: Based on the airport target area information and target time window, select airport area data that meets the spatiotemporal conditions from the airport area dataset as reference area data and comparison area data, respectively.

[0024] In some embodiments, the executing entity (e.g., a computing device) of the airport anomaly area identification method can select airport area data that meets spatiotemporal conditions from the airport area dataset as reference area data and comparison area data, respectively, based on airport target area information and a target time window. The aforementioned airport target area information can be vector data representing the boundary range of the target area, including elevation values, and composed of multiple coordinates forming a vector polygon. The aforementioned target area can be the airport area requiring anomaly identification. The aforementioned airport target area information corresponds to an airport area identifier to represent the corresponding airport area. Figure 2 The diagram showing the airport area division illustrates different building types represented by different shaped icons. Before step 101, airport area 201 can be divided into multiple sub-areas based on function (e.g., runways, aprons, etc. for flight areas; terminals, jet bridges, etc. for service areas) or building type (e.g., control towers, radar towers, etc. for high-rise buildings; terminals, hangars, etc. for enclosed buildings). Each red dashed box in the diagram marks a sub-area, and each sub-area is numbered to obtain the airport area identifier representing the corresponding airport sub-area. The target time window can consist of two time points used to select specific airport area data. For example, the target time window can be (t1, t2), where time point t1 is the start time and time point t2 is the end time. The airport area dataset can consist of airport area data collected at different time points for different airport sub-areas. Each airport area data includes regional remote sensing imagery and regional raster elevation maps. Each airport area data corresponds to an airport area identifier and a data collection time. The airport area identifier can be a string representing a specific area within the airport area. The data acquisition time mentioned above can be the actual acquisition time of the corresponding regional remote sensing image and regional raster elevation map. The regional remote sensing image can be optical remote sensing image data covering the corresponding airport sub-region, used to reflect the surface appearance of the corresponding airport sub-region at a certain point in time. The regional raster elevation map can be elevation data spatially corresponding to the corresponding regional remote sensing image, used to describe the actual surface or feature height at various locations within the corresponding region. Specifically, the regional remote sensing image can be a two-dimensional raster image of RGB type, where each pixel corresponds to a spatial location and records the spectral information (i.e., RGB information) of that location. The regional raster elevation map can be DSM (Digital Surface Model) data. The spatiotemporal conditions mentioned above can be that the airport area identifier corresponding to the reference area data and the comparison area data is the same as the airport area identifier corresponding to the above-mentioned airport target area information, and the data acquisition time corresponding to each is closest to the start and end times included in the above-mentioned target time window.

[0025] In practice, the aforementioned implementing entity can acquire airport area datasets via wired or wireless connections. Then, it can select two airport area datasets that meet the aforementioned spatiotemporal conditions from the dataset, using one as the reference area dataset and the other as the comparison area dataset. Specifically, the airport area dataset with the earlier acquisition time is used as the reference area dataset, and the one with the later acquisition time is used as the comparison area dataset. This allows for comparison and identification of the later-acquired airport area dataset using the earlier-acquired data as a reference.

[0026] Furthermore, for the same airport area or sub-area, the acquisition of regional remote sensing imagery and regional raster elevation maps should be consistent or alignable in terms of spatial reference, coverage, and resolution to ensure comparability and accuracy in subsequent anomaly identification. Specifically, regional remote sensing imagery and regional raster elevation maps should use the same or convertible spatial reference system, including: the same coordinate system (e.g., the same projected coordinate system or geographic coordinate system) and the same datum (e.g., the same elevation datum). During the data acquisition or selection phase, regional remote sensing imagery should cover the corresponding airport sub-area, and the acquisition angle should be kept as consistent as possible each time, while regional raster elevation maps should cover the same or larger spatial area. In terms of resolution, regional remote sensing imagery and regional raster elevation maps should ideally be selected with the same or similar resolution.

[0027] Step 102: Register and correct the remote sensing images of the reference area and the contrast area to generate corrected reference area images and corrected contrast area images.

[0028] In some embodiments, the execution entity can perform registration and correction on the reference area remote sensing image and the contrast area remote sensing image to generate a corrected reference area image and a corrected contrast area image. The reference area remote sensing image can be a remote sensing image of the region included in the reference area data. The contrast area remote sensing image can be a remote sensing image of the region included in the contrast area data. In practice, the execution entity can perform feature point registration on the reference area remote sensing image and the contrast area remote sensing image using the feature matching algorithm to obtain a set of registered feature points. Then, the execution entity can generate a transformation matrix between the reference area remote sensing image and the contrast area remote sensing image using the registered feature point set, and perform inter-image coordinate transformation on the contrast area remote sensing image using the generated transformation matrix to obtain the corrected contrast area image, and determine the reference area remote sensing image as the corrected reference area image.

[0029] As an example, the feature matching algorithm mentioned above could be SIFT (Scale-Invariant Feature Transform), AKAZE (Accelerated-KAZE), or ORB (Oriented FAST and Rotated BRIEF).

[0030] In some optional implementations of certain embodiments, the aforementioned implementing entity may perform registration and correction of the reference area remote sensing image and the contrast area remote sensing image through the following steps:

[0031] The first step involves extracting local feature points from both the reference area and the comparison area remote sensing images, resulting in a reference feature point set and a comparison feature point set. Reference feature points can be pixel coordinates in the reference area remote sensing image, and comparison feature points can also be pixel coordinates in the comparison area remote sensing image. In practice, the execution entity can first perform RGB channel preprocessing on both the reference area and the comparison area remote sensing images (e.g., Y=0.299R+0.587G+0.114B, where Y is the pixel brightness value), converting the images to grayscale. Then, the execution entity can use the feature matching algorithm to extract local features from the processed reference area and comparison area images, obtaining the reference feature point set and the comparison feature point set, respectively.

[0032] The second step is to perform feature point registration on the aforementioned reference feature point set and comparison feature point set to obtain a set of registered feature point groups. In practice, for each reference feature point, the aforementioned execution entity can determine the corresponding (i.e., nearest neighbor) comparison feature point in the aforementioned comparison feature point set using a nearest neighbor algorithm (such as the KD-Tree algorithm), and determine the aforementioned reference feature point and the corresponding comparison feature point as the registration feature point group.

[0033] The third step involves shaping the aforementioned set of reference feature points to obtain a set of reference feature point groups. In practice, for each reference feature point, the executing entity can use the nearest neighbor algorithm (e.g., the KD-Tree algorithm) to determine the first and second nearest neighbor reference feature points corresponding to that reference feature point in the reference feature point set, and then group these three together as a reference feature point group. Thus, a reference feature point group can form a triangular grid on the corresponding reference area remote sensing image.

[0034] The fourth step involves filtering structural points from the aforementioned set of reference feature points to update the registration feature point set. In practice, for each registration feature point group in the aforementioned set, the executing entity first determines the first and second nearest neighbor comparison feature points in the aforementioned set of comparison feature points for each comparison feature point included in the registration feature point group, and identifies these three as a comparison feature point group. Then, the executing entity determines whether the remaining four feature points in the comparison feature point group corresponding to the comparison feature point and the reference feature point group corresponding to the reference feature point are registered (i.e., whether the other two reference feature points and the two comparison feature points can form two registration feature point groups that already exist in the registration feature point set). If they cannot be registered, the aforementioned registration feature point group is removed from the aforementioned registration feature point set. If registration is successful, the executing entity can determine the side lengths of the two triangles represented by the comparison feature point set and the reference feature point set using Euclidean distance, and average the ratios of the three corresponding sides of the two triangles (which can be determined by the three registration feature point sets) to obtain the consistency determination result. Finally, in response to determining that the above consistency determination result is less than or equal to the consistency threshold (e.g., 0.65), the executing entity removes the above registration feature point set from the above registration feature point set.

[0035] Because triangles are rigid, meaning their shape doesn't change without external force, creating a triangular grid in the image ensures consistent deformation in local areas. If the proportions or angles of two corresponding triangles differ significantly between the reference and contrasting remote sensing images, feature point misalignment can be accurately detected, thus eliminating incorrect feature point matches and ensuring more precise alignment between the reference and contrasting remote sensing images.

[0036] The fifth step involves transforming and correcting the aforementioned reference area remote sensing image and the aforementioned contrast area remote sensing image based on the updated registration feature point set, to generate corrected reference area image and corrected contrast area image. In practice, the aforementioned registration feature point set can be used to perform image coordinate transformation and radiometric correction on the aforementioned reference area remote sensing image and the aforementioned contrast area remote sensing image to generate corrected reference area image and corrected contrast area image.

[0037] In some optional implementations of certain embodiments, the aforementioned execution entity may perform transformation and correction on the aforementioned reference area remote sensing image and the aforementioned comparison area remote sensing image based on the updated registration feature point set through the following steps:

[0038] The first step is to generate an image transformation matrix based on the updated set of registration feature points. In practice, the aforementioned execution entity can use the RANSAC (Random Sample Consensus) algorithm to estimate the transformation matrix between each contrast feature point and each reference feature point included in the aforementioned set of registration feature points. It then uses the least squares method to iterate and obtain the transformation matrix with the minimum total error (i.e., the minimum sum of reprojection errors for all registration feature point groups) as the image transformation matrix. This allows the contrast and reference feature points included in each registration feature point group to be mutually convertible through the image transformation matrix.

[0039] The second step involves performing image transformation on the aforementioned remote sensing image of the comparison area based on the image transformation matrix, resulting in the transformed comparison area image. In practice, the image transformation matrix can be used to transform the coordinates of each pixel in the aforementioned remote sensing image of the comparison area to obtain the transformed comparison area image.

[0040] The third step is to generate an image difference map based on the transformed contrast area image and the reference area remote sensing image. In practice, the image difference map can be generated by subtracting two pixels with the same image coordinates in the transformed contrast area image and the reference area remote sensing image channel by channel.

[0041] The fourth step involves filtering stable feature points from the aforementioned image difference map to obtain a set of stable feature points. In practice, the executing entity determines the pixel difference values ​​(sum of the three channels) in the aforementioned image difference map as the pixel difference value threshold, and identifies the coordinates of each pixel less than or equal to the aforementioned pixel difference value threshold as the set of stable feature points.

[0042] Fifth, based on the aforementioned set of stable feature points, radiometric correction is performed on the transformed contrast region image to generate a corrected reference region image and a corrected contrast region image. In practice, firstly, for each pixel channel (i.e., R channel, G channel, B channel) in the transformed contrast region image, the aforementioned execution entity iteratively fits the correction coefficient α of the corresponding channel on the aforementioned set of stable feature points using the following expression and least squares method. C β C :I C t2 (x, y) = α C ×I C t1 (x, y) + β C In this context, the character 'C' represents a channel, which can characterize the R, G, and B channels. α C β CThese can be the correction coefficient and bias value corresponding to the channel represented by character C. (x, y) can be the coordinates of stable feature points in the above set of stable feature points. I represents the image, with the subscripts t1 and t2 distinguishing between the reference area remote sensing image and the transformed contrast area image. The above I... C t2 (x, y) can be the pixel value in the C channel at coordinates (x, y) in the image of the transformed contrast region. The above I C t1 (x, y) can be the pixel value in the C channel of the remote sensing image of the reference area with coordinates (x, y). Then, the aforementioned execution entity can use the correction coefficients corresponding to different channels (e.g., the correction coefficient α in the R channel). R β R The aforementioned transformed comparison area image is then subjected to channel-by-channel pixel value correction to generate a corrected reference area image. Finally, the aforementioned executing entity can determine the aforementioned reference area remote sensing image as the corrected reference area image.

[0043] Step 103: Perform regional shadow detection on the correction reference area image and the correction contrast area image to generate shadow information for each region.

[0044] In some embodiments, the execution entity can perform region shadow detection on the aforementioned corrected reference region image and the aforementioned corrected contrast region image to generate region shadow information for each region. Each region shadow information includes a region shadow probability map and shadow constraint information. In practice, the execution entity can generate a region shadow probability map using a pre-built region shadow detection model and determine relevant attributes of the region shadow probability map (e.g., determining the shadow area, shadow bounding box coordinates, and shadow area proportion of the region shadow probability map using a preset probability threshold) as negative constraint information. The region shadow detection model can be a semantic segmentation model that takes a remote sensing image as input and a region shadow probability map as output. For example, the region shadow detection model can be a semantic segmentation model. For example, the region shadow detection model can be a U-Net model, a DeepLabV3+ model, or a SegFormer model.

[0045] In some optional implementations of certain embodiments, the execution entity may perform region shadow detection on the above-mentioned corrected reference region image and the above-mentioned corrected contrast region image through the following steps:

[0046] The first step is to normalize the above-mentioned correction reference region image to obtain a normalized reference region image. In practice, the above-mentioned correction reference region image can be normalized channel by channel ((pixel value of pixel coordinates in the corresponding channel - minimum pixel value in that channel / (maximum pixel value in that channel - minimum pixel value in that channel)) to obtain a normalized reference region image.

[0047] The second step is to input the normalized reference area image into the pre-built regional shadow detection model to generate a regional shadow probability map.

[0048] In practice, many dark-colored areas exist in actual airport remote sensing images, and shadow boundaries are prone to drift with changes in imaging time and lighting conditions. Existing general shadow segmentation models often rely on color or brightness statistics to segment dark areas, making it easier to misclassify dark features as shadows and introducing unstable boundary errors in multi-temporal comparisons. To address this, this application adopts a custom shadow detection model for airport remote sensing scenarios. By introducing shadow physical priors, boundary enhancement, and multi-scale feature modeling, the model can more accurately distinguish between real shadows and dark-colored materials, and improve the consistency of shadow boundaries in multi-temporal images. This effectively reduces the interference of shadow changes on the identification of abnormal areas and improves the reliability and stability of the overall detection results.

[0049] The aforementioned region shadow detection model can be composed of a physical prior layer 301, a backbone feature extraction network 302, a multi-scale shadow context module 303, a boundary enhancement branch 304, a decoder 305, and a shadow probability output head 306.

[0050] The aforementioned physical prior layer 301 consists of a brightness normalization module 3011 and a texture prior module 3012, used to characterize the brightness variation characteristics and texture preservation characteristics of shadow areas relative to non-shadow areas from a physical imaging perspective. The brightness normalization module 3011 first performs brightness calculations pixel-by-pixel on the input normalized region image to convert the multi-channel image into a single-channel brightness representation. For example, for an RGB image, the brightness channel value Y at each pixel location can be calculated using the formula Y = 0.299R + 0.587G + 0.114B. Here, R, G, and B represent the red, green, and blue channel pixel values ​​at the corresponding pixel location, respectively, and Y represents the brightness value at that pixel location. Subsequently, the brightness normalization module 3011 performs mean filtering on the brightness channel value Y with a preset window size (e.g., 15×15) centered on each pixel to obtain the local average brightness value within the pixel's neighborhood. The difference between this local average brightness value and the original brightness value is used to generate a relative darkness map D. The relative darkness map D is used to characterize the degree of decrease in brightness at each pixel location relative to its local neighborhood. The relative darkness map D can be represented as the normalized brightness difference; the larger the value, the darker the pixel location is relative to the surrounding area.

[0051] The texture prior module 3012 described above first takes a relative darkness map D as input and performs gradient calculations to characterize the spatial continuity of brightness changes. Specifically, the Sobel operator can be used to calculate the gradient components of the relative darkness map D in the horizontal and vertical directions, and further calculate the gradient magnitude at each pixel location to obtain a gradient magnitude map G. Each gradient magnitude reflects the intensity of the spatial change in relative darkness. Subsequently, the texture prior module 3012 can generate a relative texture map T from the gradient magnitude map G. The relative texture map T is used to characterize the degree of texture preservation at pixel locations. Specifically, the gradient magnitude map G can be normalized so that its value reflects whether the brightness change has a continuous texture structure: when a certain area darkens overall but still maintains a relatively obvious gradient change, its corresponding relative texture value is larger; when a certain area presents a large area of ​​uniform dark color and a weak gradient change, its corresponding relative texture value is smaller. Thus, the relative texture map T can be used to distinguish between "brightness decrease caused by shadows" and "low brightness areas caused by the inherent darkness of the material". After obtaining the relative darkness map D, the relative texture map T, and the brightness channel values ​​Y at each pixel location, the texture prior module 3012 can concatenate these three types of features along the channel dimension to form a prior feature tensor Fprior. Therefore, the prior feature tensor Fprior simultaneously contains brightness intensity information, relative darkness information, and texture preservation information. Finally, the texture prior module 3012 can perform feature mapping and dimensionality enhancement on the prior feature tensor Fprior through at least one 3×3 convolutional layer, a normalization layer, and the SiLU activation function, converting it into a prior feature representation Fprior′ with a preset number of channels (e.g., 32 channels) to serve as one of the inputs to the subsequent backbone feature extraction network.

[0052] The aforementioned backbone feature extraction network 302 is used for multi-level feature extraction of the input image. Its overall structure consists of four sequentially connected downsampling layers: a first downsampling layer 3021, a second downsampling layer 3022, a third downsampling layer 3023, and a fourth downsampling layer 3024, used to extract image features at different spatial scales step by step. The first downsampling layer 3021 can be composed of a convolutional layer with a kernel size of 3×3 and a stride of 2, a normalization layer, and a SiLU activation function connected sequentially. It is used to perform initial downsampling of the input image and extract shallow features, thereby outputting the first-layer feature E1. The structures of the second downsampling layer 3022, the third downsampling layer 3023, and the fourth downsampling layer 3024 can be identical, each composed of a preset number of residual blocks connected sequentially, to enhance feature representation while reducing spatial resolution. Each downsampling layer outputs features E2, E3, and E4 at the corresponding level. As an example, the number of residual blocks N can be set to 4, but this is not specifically limited here. Furthermore, to effectively incorporate the prior feature representation Fprior′ output by the aforementioned physical prior layer 301 into the backbone feature extraction process, the backbone feature extraction network 302 can also perform channel mapping on the aforementioned prior feature representation Fprior′ through a 1×1 convolutional layer, making its number of channels consistent with the number of output feature channels of the first downsampling layer 3021. Subsequently, the mapped prior feature representation Fprior′ is fused with the aforementioned first-layer feature E1 (e.g., added) to obtain the fused feature E1′. Thus, by introducing prior features in the shallow stages of the backbone network, the aforementioned backbone feature extraction network 302 can integrate shadow-related physical prior information into the subsequent feature expression process while maintaining multi-level feature extraction capabilities, thereby providing more stable and discriminative basic features for subsequent shadow detection and feature fusion.

[0053] The aforementioned multi-scale shadow context module 303 is used to acquire shadow context information at different spatial scales based on high-level semantic features, thereby enhancing the network's ability to perceive large-scale shadow regions and slender shadow structures. The input to the multi-scale shadow context module 303 can be the fourth-layer feature E4 output by the backbone feature extraction network 302. The multi-scale shadow context module 303 can have multiple context extraction branches set in parallel, each branch used to extract features from the fourth-layer feature E4 at different receptive field scales, including a first context branch 3031, a second context branch 3032, a third context branch 3033, and a fourth context branch 3034.

[0054] The first context branch 3031 can be composed of a convolutional layer with a kernel size of 3×3 and a dilation rate of 1, a normalized layer, and a SiLU activation function connected sequentially. It is used to extract shadow context features within a local range and outputs feature C1. The second context branch 3032 can be composed of a convolutional layer with a kernel size of 3×3 and a dilation rate of 3, a normalized layer, and a SiLU activation function connected sequentially. It is used to extract shadow context features within a medium-scale range and outputs feature C3. The third context branch 3033 can be composed of a convolutional layer with a kernel size of 3×3 and a dilation rate of 6, a normalized layer, and a SiLU activation function connected sequentially. It is used to extract shadow context features within a larger scale range and outputs feature C6. The fourth context branch 3034 can be composed of a convolutional layer with a kernel size of 3×3 and a dilation rate of 9, a normalized layer, and a SiLU activation function connected sequentially. It is used to extract shadow context features within a larger receptive field and outputs feature C9. Each context branch, while maintaining the input feature space resolution, achieves parallel modeling of information at different spatial scales through different dilation rates. Subsequently, the multi-scale shadow context module 303 concatenates the features (C1, C3, C6, and C9) output by the aforementioned context branches along the channel dimension to form a multi-scale context feature tensor C. Further, the multi-scale shadow context module 303 performs channel compression and feature fusion on the multi-scale context feature tensor C using a 1×1 convolutional layer to obtain the fused shadow context feature C′. Thus, the multi-scale shadow context module 303 can simultaneously model shadow distribution characteristics at different spatial scales on a single feature layer, thereby improving the network's comprehensive perception ability of large-area building shadows, slender facility shadows, and complex shadow morphologies in airport scenes.

[0055] The aforementioned boundary enhancement branch 304 is used to explicitly model the boundary information of the shadow region, thereby enhancing the network's ability to perceive the position of the shadow edge and improving the spatial stability of the shadow detection results in the boundary region. The boundary enhancement branch 304 can be composed of a first boundary feature extraction unit 3041 and a second boundary feature extraction unit 3042. Its input can be the mid-to-shallow layer features (i.e., features E2 and E3) output by the backbone feature extraction network 302, used to balance boundary detail information with semantic information at a certain level. Specifically, the first boundary feature extraction unit 3041 can take the second layer feature E2 output by the backbone feature extraction network 302 as input, and is composed of at least one 3×3 convolutional layer, a normalization layer, and a SiLU activation function connected sequentially, used to extract boundary detail features at relatively high resolution, thus obtaining the first boundary feature B2. The second boundary feature extraction unit 3042 can take the third-layer feature E3 output by the backbone feature extraction network 302 as input, and can also be composed of at least one convolutional layer with a kernel size of 3×3, a normalization layer, and a SiLU activation function connected in sequence to extract boundary-related features with certain semantic information. Subsequently, the second boundary feature extraction unit 3042 can upsample the extracted boundary features to make their spatial resolution consistent with the first boundary feature B2, thereby obtaining the second boundary feature B3.

[0056] After obtaining the first boundary feature B2 and the second boundary feature B3, the boundary enhancement branch 304 concatenates them along the channel dimension to form a fused boundary feature. This fused boundary feature is then integrated using at least one 3×3 convolutional layer to enhance the consistency between boundary information at different levels. Finally, the boundary enhancement branch 304 performs channel mapping on the integrated boundary feature using a 1×1 convolutional layer and generates a boundary probability map Pbd using a Sigmoid activation function. The boundary probability map Pbd represents the probability that each pixel location belongs to a shadow boundary. Thus, the boundary enhancement branch 304 can introduce semantic constraints at a certain level while maintaining high spatial resolution information, thereby improving the localization stability of shadow boundaries in multi-temporal images and providing reliable boundary guidance information for feature fusion in the subsequent decoding stage.

[0057] The decoder 305 is used to perform stepwise upsampling and feature recovery on the high-level features (C′) output by the multi-scale shadow context module 303. During the spatial resolution recovery process, it fuses multi-layer features from the backbone feature extraction network 302 to obtain decoded features that combine semantic and spatial detail information. Simultaneously, the decoder 305 also introduces boundary information (i.e., the boundary probability map Pbd) output by the boundary enhancement branch 304 during feature fusion to explicitly constrain the shadow boundary region. The decoder 305 can be composed of a first decoding unit, a second decoding unit, and a third decoding unit connected sequentially to progressively improve the spatial resolution of the features. The first decoding unit can take the fused shadow context feature C′ output by the multi-scale shadow context module 303 as input and perform upsampling on the fused shadow context feature C′ to ensure its spatial resolution is consistent with the third-layer feature E3 output by the backbone feature extraction network 302. Subsequently, the first decoding unit can fuse the upsampled features with the third-layer features E3, and integrate the features through at least one convolutional layer, a normalization layer, and the SiLU activation function to obtain the first decoded feature D3.

[0058] The second decoding unit can take the first decoding feature D3 as input and upsample it to make its spatial resolution consistent with the second-layer feature E2 output by the backbone feature extraction network 302. Subsequently, the second decoding unit can fuse the upsampled feature with the second-layer feature E2 and integrate the features through at least one convolutional layer, a normalization layer, and a SiLU activation function to obtain the second decoding feature D2.

[0059] The third decoding unit can take the second decoding feature D2 as input and upsample it to make its spatial resolution consistent with the first-layer fusion feature E1′ output by the backbone feature extraction network 302. Subsequently, the third decoding unit can fuse the upsampled second decoding feature with the first-layer fusion feature E1′, and integrate the features through at least one convolutional layer, a normalization layer, and a SiLU activation function to obtain the final decoding feature D1. Thus, through the above-mentioned step-by-step upsampling and cross-layer feature fusion method, the decoder 305 can restore the spatial resolution while introducing feature information from different layers of the backbone network, thereby balancing the overall semantic consistency of the shadow region with the ability to express boundary details.

[0060] It should be noted that each decoding unit extracts the hierarchical features (i.e., E3, E2, and E3) from the network output by combining the decoding features of this layer with the backbone features. During fusion, to constrain the decoding process by introducing shadow boundary information (i.e., the boundary probability map Pbd), each decoding unit can also acquire the boundary probability map Pbd output by the boundary enhancement branch 304, and perform downsampling or upsampling on the boundary probability map to ensure its spatial resolution is consistent with the corresponding decoding features (i.e., D3, D2, D1). Subsequently, each decoding unit can perform boundary gating on the corresponding decoding features using the boundary probability map Pbd to update the decoding features. Specifically, the boundary gating can be expressed as Dk = Dk × (1 + μ × Up(Pbd)). Here, Dk can be the decoding feature output by the corresponding decoding unit (e.g., D3, D2, or D1). Up() represents the upsampling operation, which can also be replaced with a downsampling operation to ensure the spatial resolution of the boundary probability map is consistent with the corresponding decoding features (i.e., D3, D2, D1). μ represents the enhancement coefficient, which can be 0.5.

[0061] It should be noted that the above upsampling process can be interpolation upsampling, that is, upsampling the input features through interpolation to increase their spatial resolution to the target size. This interpolation upsampling can be nearest neighbor interpolation upsampling or bilinear interpolation upsampling. When the input feature size is... The target size is Interpolation can be used to supplement pixel values ​​in the spatial dimension, thereby obtaining features with the same resolution as the target features. Here, H and W represent the number of pixels in the row and column dimensions of the input feature, respectively. The above upsampling process can also be performed by upsampling the input features using a transposed convolutional layer. For example, a convolutional kernel with a stride greater than 1 can be used to perform convolution operations on the input features, thus enlarging the output features in the spatial dimension. The above downsampling process can be convolutional downsampling. For example, using a convolutional layer with a stride of 2 can reduce the input features to half their original size in the spatial dimension. The above downsampling can also be pooling downsampling, including max pooling or average pooling.

[0062] The aforementioned shadow probability output head 306 is used to convert the final decoded feature D1 output by the decoder 305 into a pixel-level shadow probability representation to generate a region shadow probability map. The shadow probability output head 306 can take the final decoded feature D1 as input and include at least one convolutional layer with a kernel size of 1×1 to perform channel mapping on the final decoded feature D1, converting its channel count from the feature channel count to a preset output channel count. Subsequently, the shadow probability output head 306 can process the mapped feature using a Sigmoid activation function to output a shadow probability map Psh. The value at each pixel position in the shadow probability map represents the probability (i.e., probability value) that the corresponding position belongs to a shadow region, and the probability value is limited to a preset interval (e.g., [0, 1]).

[0063] The third step involves generating a regional shadow gating map as shadow constraint information based on the generated regional shadow probability map. In practice, the regional shadow probability map can be scaled (e.g., upsampled or downsampled) to ensure its spatial resolution matches the corresponding regional image master feature (i.e., the corresponding fourth-layer feature E4). Then, the shadow probability values ​​in the scaled regional shadow probability map are inverted (i.e., subtracted by one and then the absolute value is taken) to generate the regional shadow gating map. Pixels with higher shadow probability values ​​in the regional shadow probability map have smaller gating weights in the regional shadow gating map, while pixels with lower shadow probability values ​​have larger gating weights.

[0064] The fourth step is to determine the generated regional shadow probability map and shadow constraint information as the regional shadow information corresponding to the aforementioned corrected reference region image. In practice, the generated regional shadow probability map and shadow constraint information can be determined as the regional shadow information corresponding to the aforementioned corrected reference region image.

[0065] Fifth, based on the aforementioned regional shadow detection model, generate regional shadow information corresponding to the corrected comparison region image. In practice, the specific implementation method of "generating regional shadow information corresponding to the corrected comparison region image based on the aforementioned regional shadow detection model" can refer to the specific implementation methods of steps one to six above, and will not be repeated here.

[0066] Step 104: Based on the corresponding shadow probability maps of each region, perform constrained feature fusion on the corrected reference region image and the corrected contrast region image respectively to generate the main features of the reference region and the main features of the contrast region.

[0067] In some embodiments, the execution entity can perform constrained feature fusion on the corrected reference region image and the corrected contrast region image respectively, based on the region shadow probability maps corresponding to the corrected reference region image and the corrected contrast region image, to generate the main features of the reference region and the main features of the contrast region. In practice, firstly, image features can be extracted from the corrected reference region image using the aforementioned feature extraction network, and the extracted reference region image features are multiplied by the corresponding region shadow probability map to generate the main features of the reference region. Then, image features can be extracted from the corrected contrast region image using the aforementioned feature extraction network, and the extracted contrast region image features are multiplied by the corresponding region shadow probability map to generate the main features of the contrast region.

[0068] In some optional implementations of certain embodiments, the execution entity may perform constrained feature fusion on the correction reference region image and the correction contrast region image respectively based on the corresponding shadow probability maps of each region through the following steps:

[0069] The first step involves extracting basic features from the aforementioned corrected reference region image using a feature extraction network to generate the main features of the region image. This feature extraction network can reuse the aforementioned backbone feature extraction network. In practice, the corrected reference region image can be input into the backbone feature extraction network to obtain the hierarchical features E1, E2, E3, and E4 output from the first, second, third, and fourth downsampling layers of the backbone feature extraction network. The fourth layer feature, E4, is then selected as the main feature of the region image.

[0070] The second step involves extracting edge texture features from the aforementioned correction reference region image to generate edge texture features for the region image. In practice, the correction reference region image can be converted into a grayscale image, and then processed using an edge detection operator (such as the Sobel operator) to determine the gradient changes in the horizontal and vertical directions, thereby obtaining a gradient magnitude map. Finally, the execution entity can convert the gradient magnitude map into a feature representation that matches the main features of the aforementioned region image in terms of spatial scale through downsampling or 1×1 convolution operations, serving as the edge texture features of the region image.

[0071] The third step involves extracting brightness features from the aforementioned correction reference area image to generate regional image brightness features. In practice, firstly, the correction reference area image can be converted into a brightness map (e.g., by weighted summation of the pixel values ​​of the R, G, and B channels to obtain the brightness values ​​corresponding to each coordinate). Then, the executing entity can determine the average brightness and variance of the brightness values ​​within each local region (e.g., a 7×7 window) in the brightness map to form a brightness feature representation. Subsequently, the brightness features can be converted into a feature form consistent with the main features of the regional image in terms of spatial scale through downsampling or feature mapping, thereby obtaining the regional image brightness features.

[0072] The fourth step involves spatially aligning the main features, edge texture features, and brightness features of the aforementioned regional image to obtain aligned main features, aligned edge texture features, and aligned brightness features. In practice, features with inconsistent spatial resolutions (i.e., the aforementioned main features, edge texture features, and brightness features) can be upsampled or downsampled to maintain consistent spatial dimensions. Finally, a 1×1 convolutional layer is used to map the number of channels for different features, unifying the channel count to obtain aligned main features, aligned edge texture features, and aligned brightness features.

[0073] The fifth step is to generate the alignment region image sub-features based on the alignment edge texture features and alignment brightness features described above. In practice, the alignment edge texture features and alignment brightness features can be concatenated along the channel dimension, and the concatenated features can be fused using a convolution operation (e.g., a 3×3 convolution operation) to generate the alignment region image sub-features.

[0074] Step 6: Based on the shadow constraint information corresponding to the aforementioned corrected reference area image, feature constraints are applied to the secondary features of the aforementioned aligned area image to update the secondary features of the aligned area image. In practice, the shadow constraint information corresponding to the aforementioned corrected reference area image (i.e., the regional shadow gating map) can be used as a weight map to weight the feature values ​​of the secondary features of the aforementioned aligned area image at each spatial location. This reduces the weight of secondary features in areas with higher shadow probability, while maintaining or enhancing the weight of secondary features in areas with lower shadow probability, thereby updating the secondary features of the aligned area image.

[0075] Step 7: Based on the shadow constraint information corresponding to the above-mentioned corrected reference area image, feature fusion is performed on the main features of the above-mentioned aligned area image and the updated secondary features of the aligned area image to generate the main features of the reference area. In practice, the above-mentioned execution entity can perform weighted fusion on the main features of the above-mentioned aligned area image and the updated secondary features of the aligned area image. The weights are adjusted by the shadow constraint information (i.e., the regional shadow gating map) corresponding to the above-mentioned corrected reference area image, so that the secondary features of the aligned area image in the non-shadowed area have a larger weight, while the main features of the aligned area image in the shadowed area have a larger weight. The above weighted fusion operation can be expressed as: F(x, y) = Fsub(x, y) × G(x, y) + (1-G(x, y)) × Fmain(x, y). Where F(x, y) can be the feature value of the main feature of the reference area at (x, y). G(x, y) can be the value (i.e., weight) of the regional shadow gating map at (x, y). Fmain(x, y) can be the feature value (i.e., pixel value) of the main feature of the aligned area image at (x, y). Fsub(x, y) can be the feature value (i.e., pixel value) of the updated aligned region image sub-feature at (x, y). The above (x, y) can be the row and column coordinates of the aligned region image main feature or the updated aligned region image sub-feature.

[0076] Step 8: Generate the main features of the contrast region based on the above-mentioned corrected contrast region image. In practice, the specific implementation method of "generating the main features of the contrast region based on the above-mentioned corrected contrast region image" can refer to the specific implementation methods of steps 1 to 8 above, and will not be repeated here.

[0077] The aforementioned steps one through eight constitute an inventive point of this disclosure, addressing the technical problem that "during airport remote sensing anomaly detection, due to differences in lighting conditions and shadow position variations between images from different time periods, the brightness, texture, and edge features of buildings, facilities, and other ground features in remote sensing images are prone to significant changes. This leads to apparent differences caused solely by changes in lighting or shadow being misidentified as genuine anomalies, resulting in numerous false anomalies and insufficient stability in the anomaly area identification results, making it difficult to accurately reflect real safety-related changes within the airport area." Solving these factors can improve the stability of anomaly area identification and enhance airport area security. To achieve this, a constraint fusion mechanism based on shadow information is introduced during the image master feature generation process. This suppresses features susceptible to lighting effects within shadow areas and strengthens the expression of real ground feature structures within non-shadow areas. This allows for the generation of more stable and comparable regional master features between remote sensing images from different times, effectively reducing false anomalies caused by shadow changes, improving the accuracy of airport anomaly area identification results, and enhancing airport area security.

[0078] Step 105: Identify abnormal regions by analyzing the main features of the reference region and the main features of the comparison region to generate candidate abnormal region identification information.

[0079] In some embodiments, the execution entity can perform anomaly region identification on the main features of the reference region and the main features of the comparison region to generate candidate anomaly region identification information. In practice, the execution entity can calculate the difference between the main features of the reference region and the main features of the comparison region at the same spatial location to generate a region difference map. As an example, pixel-by-pixel absolute difference can be performed on the two, and the differences of multiple channels can be summed to obtain the change intensity value at each pixel location. Then, the region difference map can be thresholded according to a preset change intensity threshold to obtain a candidate change mask. Then, connected component analysis can be performed on the candidate change mask to aggregate adjacent changed pixels into several continuous regions, and each continuous region can be determined as a candidate anomaly region range. Finally, the basic region attributes, such as region area, mean or maximum difference intensity within the region, can be calculated for each candidate anomaly region range, and the candidate anomaly region range and its region attributes can be combined to form candidate anomaly region identification information.

[0080] In some optional implementations of certain embodiments, the aforementioned execution entity may perform abnormal region identification on the main features of the reference region and the main features of the comparison region through the following steps:

[0081] The first step involves generating hierarchical difference features for each region based on the reference region hierarchical feature set corresponding to the main features of the aforementioned reference region and the comparison region hierarchical feature set corresponding to the main features of the aforementioned comparison region. Specifically, the reference region hierarchical feature set corresponding to the main features of the aforementioned reference region can be the hierarchical features output by each downsampling layer after the remote sensing image of the aforementioned reference region is input into the aforementioned backbone feature extraction network (i.e., features E1, E2, and E3 corresponding to the remote sensing image of the aforementioned reference region). Similarly, the comparison region hierarchical feature set corresponding to the main features of the comparison region can be the hierarchical features output by each downsampling layer after the remote sensing image of the comparison region is input into the aforementioned backbone feature extraction network (i.e., features E1, E2, and E3 corresponding to the remote sensing image of the aforementioned comparison region).

[0082] In practice, the aforementioned implementing entity can subtract the reference region hierarchical features belonging to the same level (i.e., originating from the same downsampling layer) from the comparison region hierarchical features element by element to obtain the hierarchical difference features of each region.

[0083] The second step involves applying shadow constraints to the hierarchical difference features of each region to update these features. In practice, a region shadow gating map with the same spatial resolution as each level can be used as a weight map (i.e., a region shadow gating map generated by aligning the region shadow probability map with the corresponding hierarchical features). This weights the regional hierarchical difference features at each level pixel-by-pixel, thus reducing the weight of difference features located within shadowed areas while maintaining or enhancing the weight of difference features located outside shadowed areas. This method reduces spurious differences caused by illumination variations within shadowed areas.

[0084] The third step is to aggregate the cumulative differences in the updated regional hierarchical features to generate a regional cumulative difference change map. In practice, the difference features at each level can be accumulated pixel by pixel or weighted summed to generate a regional cumulative difference change map. In the regional cumulative difference change map, regions with larger values ​​indicate that there are significant changes in features at multiple levels at that location.

[0085] The fourth step involves identifying candidate regions from the cumulative difference change map to generate information on the range of each anomalous region. In practice, the executing entity can use a preset change intensity threshold to perform threshold segmentation on the cumulative difference change map, filtering out pixel regions with change intensity greater than the preset threshold. Then, the filtered, spatially adjacent pixel regions can be merged into connected regions to obtain several continuous change regions. Each continuous change region is defined as an anomalous region range, and the pixel coordinates representing the anomalous region range are determined as the corresponding anomalous region range information.

[0086] It should be noted that each anomalous area information can be characterized in remote sensing images as a corresponding continuous area exhibiting significantly different change characteristics from the surrounding area between the start and end time points. These changes can refer to the appearance of new structures or objects in locations where no features were previously present, significant changes in the shape, boundaries, or outlines of existing features, or a continuous alteration in the texture and brightness distribution within the area, and these changes are not caused by illumination or shadows. In specific application scenarios, this might represent the presence of newly added or expanded buildings within the corresponding airport area, an increase in the height of ground-based objects, equipment, or structures (e.g., protruding foreign objects on the apron), or changes in features that may encroach on airspace restrictions. If these changes occur near runways, taxiways, or airspace-sensitive areas, they may pose a significant risk to flight safety, navigation safety, or airport operational safety, and therefore require timely identification.

[0087] The fifth step involves extracting regional attributes from the candidate anomaly regions represented by each anomaly region's extent information to generate anomaly region attributes. In practice, the attributes can include the area of ​​the anomaly region corresponding to each anomaly region's extent information, the average value of the cumulative difference change map within the region, and the maximum and minimum values ​​of the cumulative difference change map within the region. These anomaly region attributes are used to quantitatively describe the changes within the anomaly region's extent, providing supplementary information on the "degree of change" and "spatial characteristics" of the anomaly region.

[0088] The sixth step is to determine the generated range information of each abnormal region and the corresponding attributes of each abnormal region as candidate abnormal region identification information.

[0089] Step 106: Based on the airport target area information, the comparison area raster elevation map, the reference area raster elevation map, and various shadow constraint information, the candidate abnormal area identification information is verified and graded to obtain the airport target area abnormal information.

[0090] In some embodiments, the executing entity can verify and classify the candidate abnormal region identification information based on the airport target area information, the comparison area raster elevation map, the reference area raster elevation map, and various shadow constraint information to obtain airport target area abnormal information. In practice, firstly, the executing entity can determine the height restriction conditions corresponding to the target area according to the airport target area information and use the height restriction conditions as height verification values. Then, the elevation values ​​at the same raster position in the comparison area raster elevation map and the reference area raster elevation map can be subtracted to generate a height change difference, and the elevation value in the comparison area raster elevation map can be subtracted from the height restriction conditions to obtain a height exceedance difference. Then, in response to determining that there are raster cells with height change differences or height exceedance differences greater than or equal to the exceedance threshold within the candidate abnormal region, the corresponding raster cells can be identified as exceedance cells, and the exceedance cell information can be associated with the corresponding candidate abnormal region to update the candidate abnormal region identification information. Finally, the abnormal area judgment score can be generated by combining the over-limit unit information, shadow constraint information and regional attributes of the candidate abnormal area, and the abnormal area level can be determined according to the preset grading threshold. The abnormal area range information, abnormal area judgment score, abnormal area level and regional attributes can be determined as the target area abnormal information, and the airport target area abnormal information can be obtained by summarizing them.

[0091] In some optional implementations of certain embodiments, the aforementioned execution entity may perform verification and classification of the aforementioned candidate abnormal region identification information through the following steps:

[0092] The first step is to determine the regional height restriction information based on the aforementioned airport target area information. In practice, the maximum permissible height value (i.e., the maximum elevation value) within the area represented by the aforementioned airport target area information can be determined as the regional height restriction information based on the vectorized data contained in the aforementioned airport target area information.

[0093] The second step involves generating the height variation differences for each location based on the aforementioned raster elevation maps of the comparison and reference areas. In practice, this can be achieved by subtracting the raster elevation maps of the reference and comparison areas point by point at the same grid location to obtain the height variation differences for each position.

[0094] The third step is to verify the height of the comparison area raster elevation map based on the aforementioned regional height restriction information, obtaining the height differences for each value. In practice, the elevation values ​​of each grid position in the comparison area raster elevation map can be subtracted from the aforementioned regional height restriction information, and negative values ​​can be removed to obtain the height differences for each value.

[0095] The fourth step involves generating at least one out-of-limit unit information in response to the determination that there is a height difference or height change difference greater than or equal to the out-of-limit threshold among the various height differences and various height change differences. In practice, when it is determined that there is a height change difference or a height difference greater than or equal to the preset out-of-limit threshold, at least one corresponding grid can be identified as at least one out-of-limit unit, and the position coordinates corresponding to at least one grid can be identified as at least one out-of-limit unit information.

[0096] The fifth step is to associate the generated out-of-limit unit information with the aforementioned candidate anomaly region identification information to update the candidate anomaly region identification information. In practice, it can be determined whether the location coordinates corresponding to each out-of-limit unit are within the range of each candidate anomaly region represented by the aforementioned candidate anomaly region identification information. If they are within the range, the corresponding out-of-limit unit information is associated with the corresponding anomaly region or anomaly region range information.

[0097] The sixth step is to classify the updated candidate anomaly area identification information and generate anomaly information for the airport target area.

[0098] In some optional implementations of certain embodiments, the aforementioned execution entity may classify the updated candidate anomaly region identification information through the following steps:

[0099] The first step is to perform the following steps for each anomaly region range information in the updated candidate anomaly region identification information:

[0100] The first sub-step involves generating an anomaly region judgment score and an anomaly region level based on the over-limit unit information, corresponding shadow constraint information, and anomaly region attributes corresponding to the aforementioned anomaly region range information. In practice, firstly, in response to determining that the aforementioned anomaly region range information contains at least one corresponding over-limit unit information, the high-risk quantity Sh can be set to 1. Then, the executing entity can determine the change intensity quantity Sc as the ratio of the average value of the cumulative difference change map within the region included in the aforementioned anomaly region attributes to the difference between the maximum and minimum values ​​of the cumulative difference change map within the region. Subsequently, the ratio of the number of shadow pixels within the anomaly region range represented by the aforementioned anomaly region range information to the total number of pixels within the anomaly region range can be determined, and the absolute value obtained by subtracting 1 from it can be determined as the shadow confidence measure Ss. Finally, the executing entity can determine the anomaly region judgment score Score by weighted summation: Score = wh × Sh + wc × Sc + ws × Ss. Where wh, wc, and ws are preset weights, which can be 0.5, 0.3, and 0.2, respectively. Finally, in response to determining that the above-mentioned abnormal area judgment score is greater than or equal to a first score threshold (e.g., 0.7), the abnormal area level is determined to be high-risk. In response to determining that the above-mentioned abnormal area judgment score is less than the first score threshold but greater than or equal to a second score threshold (e.g., 0.4), the abnormal area level is determined to be medium-risk. In response to determining that the above-mentioned abnormal area judgment score is less than the second score threshold, the abnormal area level is determined to be low-risk.

[0101] It should be noted that the high-risk level indicates an abnormal area with a significant increase in height or exceeding the airport's height limit. This change, in remote sensing imagery, exhibits stable, continuous, and non-shadow-induced structural characteristics, posing a direct or potential threat to airport operations or flight safety. In remote sensing imagery, this may manifest as newly constructed or significantly taller structures, or as height-exceeding cells in the corresponding area on the raster elevation map. In application scenarios, this may include, but is not limited to, the following: newly constructed factories, tower cranes, or communication towers within the runway extension or airspace protection zone, or temporary but relatively tall construction facilities near the runway.

[0102] The aforementioned medium-risk level indicates significant spatial or structural changes within the corresponding abnormal area, but it has not yet been clearly determined to constitute height exceeding limits, or the height change is close to the threshold, posing a potential or developing risk to airport safety. In remote sensing imagery, this may manifest as new or expanding structures within the area, but with limited height changes, or as a trend of increasing height on the raster elevation map, but without significant exceeding of limits. In application scenarios, this may include, but is not limited to, the following: long-term stockpiled materials gradually accumulating with a continuing upward trend, or new facilities added to the apron or cargo area, but whose height does not exceed the height limit.

[0103] The aforementioned low-risk level indicates that although changes were detected within the corresponding abnormal area, the scale and intensity of the changes were small, or the changes were mainly located within the shadowed area, and the impact on airport safety was low or negligible. In remote sensing imagery, this may manifest as a small area of ​​change, or changes concentrated in the shadowed area. In application scenarios, this may include, but is not limited to, the following situations: temporary parking of small items or equipment, or local image differences caused by changes in lighting conditions.

[0104] The second sub-step involves determining the above-mentioned abnormal area range information, abnormal area judgment score, abnormal area level, and abnormal area attributes as the target area abnormal information.

[0105] The second step is to identify the generated abnormal information in each target area as abnormal information in the airport target area.

[0106] Therefore, by combining airport target area information, multi-temporal regional raster elevation maps, and shadow constraint information, the height of candidate abnormal areas can be verified to determine whether there is an excessive height or a trend of height change. Based on this, the abnormal areas can be classified into risk levels, thereby improving the stability of abnormal area identification and thus improving the airspace safety within the airport area.

[0107] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an airport anomaly area identification device, which are similar to... Figure 1 Corresponding to the method embodiments shown, the airport anomaly area identification device can be specifically applied to various electronic devices.

[0108] like Figure 4As shown, an airport anomaly area identification device 400 in some embodiments includes: a selection unit 401, a registration and correction unit 402, a region shadow detection unit 403, a constraint feature fusion unit 404, an anomaly area identification unit 405, and a verification and grading unit 406. The selection unit 401 is configured to select airport area data that meets spatiotemporal conditions from an airport area dataset as reference area data and comparison area data, respectively, based on airport target area information and a target time window. Each airport area data includes a regional remote sensing image and a regional raster elevation map. The registration and correction unit 402 is configured to perform registration and correction on the reference area remote sensing image and the comparison area remote sensing image to generate a corrected reference area image and a corrected comparison area image. The reference area remote sensing image is a regional remote sensing image included in the reference area data, and the comparison area remote sensing image is a regional remote sensing image included in the comparison area data. The region shadow detection unit 403 is configured to perform region shadow detection on the corrected reference area image and the corrected comparison area image. The system performs domain shadow detection to generate shadow information for each region, where each region's shadow information includes a region shadow probability map and shadow constraint information. A constraint feature fusion unit 404 is configured to perform constraint feature fusion on the corrected reference region image and the corrected comparison region image based on the corresponding region shadow probability maps, to generate main features for the reference region and main features for the comparison region. Anomaly region identification unit 405 is configured to identify anomalies in the main features of the reference region and the comparison region, to generate candidate anomaly region identification information. A verification and grading unit 406 is configured to verify and grade the candidate anomaly region identification information based on the comparison region raster elevation map, the reference region raster elevation map, and the shadow constraint information, to obtain anomaly information for the airport target area.

[0109] It is understandable that the units recorded in the airport abnormal area identification device 400 are related to the reference Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the airport anomaly area identification device 400 and the units contained therein, and will not be repeated here.

[0110] The following is for reference. Figure 5 It shows a schematic diagram of the structure of an electronic device 500 (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0111] like Figure 5As shown, the electronic device 500 may include a processing unit 501 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0112] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0113] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 509, or installed from storage device 508, or installed from ROM 502. When the computer program is executed by processing device 501, it performs the functions defined in the methods of some embodiments of this disclosure.

[0114] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0115] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0116] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: select airport area data that meets spatiotemporal conditions from the airport area dataset, based on airport target area information and a target time window, as reference area data and comparison area data respectively, wherein each airport area data includes regional remote sensing imagery and regional raster elevation map; and perform registration and correction on the reference area remote sensing imagery and the comparison area remote sensing imagery to generate corrected reference area imagery and corrected comparison area imagery, wherein the aforementioned reference area remote sensing imagery is the regional remote sensing imagery included in the aforementioned reference area data, and the aforementioned comparison area remote sensing imagery is the regional remote sensing imagery included in the aforementioned comparison area data. The process involves: performing region shadow detection on the aforementioned corrected reference area image and the aforementioned corrected comparison area image to generate shadow information for each region, wherein each region shadow information includes a region shadow probability map and shadow constraint information; performing constraint feature fusion on the aforementioned corrected reference area image and the aforementioned corrected comparison area image based on the corresponding region shadow probability maps to generate reference region main features and comparison region main features; identifying abnormal regions on the aforementioned reference region main features and comparison region main features to generate candidate abnormal region identification information; and verifying and classifying the aforementioned candidate abnormal region identification information based on the comparison area raster elevation map, the reference area raster elevation map, and each shadow constraint information to obtain airport target area abnormal information.

[0117] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0118] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0119] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0120] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

[0121] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0123] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0124] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for identifying abnormal areas in an airport, characterized in that, include: Based on the airport target area information and target time window, airport area data that meet the spatiotemporal conditions are selected from the airport area dataset as reference area data and comparison area data, respectively. Each airport area data includes regional remote sensing imagery and regional raster elevation map. The remote sensing images of the reference area and the remote sensing images of the comparison area are registered and corrected to generate corrected reference area images and corrected comparison area images, wherein the remote sensing images of the reference area are the regional remote sensing images included in the reference area data, and the remote sensing images of the comparison area are the regional remote sensing images included in the comparison area data. Regional shadow detection is performed on the correction reference region image and the correction contrast region image to generate shadow information for each region. Each region shadow information includes a region shadow probability map and shadow constraint information. The process of generating shadow information for the correction reference region image and the correction contrast region image includes: The image of the correction reference region is normalized to obtain a normalized reference region image; The normalized reference region image is input into a pre-constructed region shadow detection model to generate a region shadow probability map; The region shadow probability map is scaled to ensure that its spatial resolution is consistent with the main features of the corresponding region image. The shadow probability values ​​included in the scale-transformed region shadow probability map are inverted to generate a region shadow gating map as shadow constraint information. Based on the corresponding shadow probability maps of each region, constrained feature fusion is performed on the corrected reference region image and the corrected contrast region image respectively to generate the main features of the reference region and the main features of the contrast region. Anomaly region identification is performed on the main features of the reference region and the main features of the comparison region to generate candidate anomaly region identification information; Based on the airport target area information, the comparison area raster elevation map, the reference area raster elevation map, and various shadow constraint information, the candidate abnormal area identification information is verified and graded to obtain the airport target area abnormal information.

2. The method according to claim 1, characterized in that, The process of registering and correcting the remote sensing images of the reference area and the contrast area to generate corrected reference area images and corrected contrast area images includes: Local feature points are extracted from the remote sensing images of the reference area and the remote sensing images of the comparison area to obtain a reference feature point set and a comparison feature point set; The reference feature point set and the comparison feature point set are registered to obtain a set of registered feature point groups; The reference feature point set is shaped to obtain a reference feature point group set; Based on the reference feature point set, structural points are filtered in the registration feature point set to update the registration feature point set; Based on the updated set of registration feature points, the remote sensing images of the reference area and the remote sensing images of the contrast area are transformed and corrected to generate corrected reference area images and corrected contrast area images.

3. The method according to claim 2, characterized in that, The step of transforming and correcting the reference area remote sensing image and the contrast area remote sensing image based on the updated registration feature point set to generate a corrected reference area image and a corrected contrast area image includes: Based on the updated set of registration feature points, generate the image transformation matrix; Based on the image transformation matrix, the remote sensing image of the comparison area is transformed to obtain the transformed comparison area image. Based on the transformed comparison area image and the reference area remote sensing image, an image difference map is generated; The image difference map is subjected to stable feature point screening to obtain a set of stable feature points; Based on the set of stable feature points, radiometric correction is performed on the transformed contrast region image to generate a corrected reference region image and a corrected contrast region image.

4. The method according to claim 1, characterized in that, The step of identifying abnormal regions by analyzing the main features of the reference region and the main features of the comparison region to generate candidate abnormal region identification information includes: Based on the reference region hierarchical feature set corresponding to the main feature of the reference region and the comparison region hierarchical feature set corresponding to the main feature of the comparison region, generate hierarchical difference features for each region. Shading constraints are applied to the hierarchical difference features of each region to update the hierarchical difference features of each region; The updated regional hierarchical difference features are aggregated to generate a regional cumulative difference change map. Candidate regions are identified from the cumulative difference change map of the region to generate information on the range of each abnormal region; For each candidate abnormal region represented by the abnormal region range information, regional attributes are extracted to generate abnormal region attributes; The generated range information of each abnormal region and the corresponding attributes of each abnormal region are determined as candidate abnormal region identification information.

5. The method according to claim 4, characterized in that, The process involves verifying and classifying the candidate anomaly identification information based on the airport target area information, the comparison area raster elevation map, the reference area raster elevation map, and various shadow constraint information to obtain airport target area anomaly information, including: Based on the airport target area information, determine the area height restriction information; Based on the raster elevation map of the comparison area and the raster elevation map of the reference area, generate various height variation differences; Based on the height restriction information of the region, the height of the comparison area raster elevation map is verified to obtain various height differences; In response to determining that there is a height difference or height change difference greater than or equal to the over-limit threshold among the various height differences and various height change differences, at least one over-limit unit information is generated; Associate the generated information of at least one out-of-limit unit with the candidate anomaly region identification information to update the candidate anomaly region identification information; The updated candidate anomaly region identification information is classified to generate anomaly information for the airport target area.

6. The method according to claim 5, characterized in that, The step of classifying the updated candidate anomaly region identification information to generate airport target area anomaly information includes: For each anomaly region range information in the updated candidate anomaly region identification information, perform the following steps: Based on the out-of-limit unit information, the corresponding shadow constraint information, and the abnormal region attributes corresponding to the abnormal region range information, an abnormal region judgment score and an abnormal region level are generated. The abnormal region range information, the abnormal region judgment score, the abnormal region level, and the abnormal region attributes are determined as the target region abnormal information. The generated abnormal information in each target area is identified as abnormal information in the airport target area.

7. An airport anomaly area identification device, characterized in that, include: The selection unit is configured to select airport area data that meets the spatiotemporal conditions from the airport area dataset based on the airport target area information and the target time window, respectively as reference area data and comparison area data. Each airport area data includes regional remote sensing imagery and regional raster elevation map. The registration and correction unit is configured to register and correct a reference area remote sensing image and a contrast area remote sensing image to generate a corrected reference area image and a corrected contrast area image, wherein the reference area remote sensing image is a regional remote sensing image included in the reference area data, and the contrast area remote sensing image is a regional remote sensing image included in the contrast area data. A region shadow detection unit is configured to perform region shadow detection on the corrected reference region image and the corrected contrast region image to generate region shadow information for each region. Each region shadow information includes a region shadow probability map and shadow constraint information. The process of performing region shadow detection on the corrected reference region image and the corrected contrast region image to generate region shadow information includes: normalizing the corrected reference region image to obtain a normalized reference region image; inputting the normalized reference region image into a pre-constructed region shadow detection model to generate a region shadow probability map; scaling the region shadow probability map to ensure its spatial resolution matches the main features of the corresponding region image; and inverting the shadow probability values ​​included in the scaled region shadow probability map to generate a region shadow gating map as shadow constraint information. The constraint feature fusion unit is configured to perform constraint feature fusion on the corrected reference region image and the corrected contrast region image respectively according to the corresponding shadow probability maps of each region, so as to generate the main feature of the reference region and the main feature of the contrast region. An abnormal region identification unit is configured to identify abnormal regions by analyzing the main features of the reference region and the main features of the comparison region, so as to generate candidate abnormal region identification information. The verification and grading unit is configured to verify and grade the candidate abnormal area identification information based on the airport target area information, the comparison area raster elevation map, the reference area raster elevation map, and various shadow constraint information, so as to obtain the airport target area abnormal information.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images