Line merging method and system based on airport regulations AIP diagram
Through deep contour perception network and probability calculation technologies, the problem of incomplete and overlapping processing of line extraction in AIP graphs is solved, the accuracy and efficiency of line graphs are improved, manual intervention is reduced, and the accuracy of airport digital management is supported. and efficiency.
Patent Information
- Application Number
- CN202411117085.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-15
AI Technical Summary
When handling airport details AIP diagrams, the existing technology uses incomplete line extraction, improper processing of overlapping lines, and wrong attribute text correlation, resulting in low accuracy of generated line drawings, affecting the efficiency and accuracy of airport digital management.
Line merging methods based on depth contour perception networks, including image fusion, text removal and line detection, line detection and merge through feature extraction module, multi-scale edge detection module and line prediction module, combined with probability calculation and spectral clustering algorithm processing Overlap lines and associate attribute text through visual attention mechanisms.
Improve the accuracy and efficiency of line extraction and merging, reduce manual intervention, improve the accuracy of generated line drawings, and provide more accurate and timely information support for aviation safety management and airport operations.
Smart Images

Figure CN118644515B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of image data processing, and in particular relates to a line merging method and system based on an airport specification AIP diagram. Background Art
[0002] In the aviation field, Aeronautical Information Publication (AIP) diagrams are vital reference materials that record in detail the physical structure, operating rules, navigation facilities and other key information of the airport. AIP diagrams are usually composed of various lines, symbols and texts, graphically displaying important elements such as the airport's runways, taxiways, aprons, terminals, obstacles, navigation points, etc. With the continuous growth of air traffic and higher requirements for airport operation efficiency and safety, digital AIP diagrams are becoming more and more important. However, manual interpretation and information extraction of AIP diagrams are not only time-consuming and labor-intensive, but also prone to errors. Therefore, it is urgent to develop a technology for automatically processing AIP diagrams.
[0003] Traditionally, the application of image processing technology in the aviation field has made some progress, such as using image segmentation, edge detection, feature matching and other methods to process satellite images or aerial photographs. However, the processing of AIP images faces unique challenges, including complex line structures, dense symbols and texts, and overlapping and interrupted lines. Existing image processing algorithms often perform poorly when processing such detail-intensive images, especially in dealing with overlapping lines, the connection of discontinuous lines, and the recognition and association of attribute text related to lines, resulting in incomplete or erroneous extracted lines, affecting the efficiency and accuracy of digital management of airports. Summary of the invention
[0004] In order to solve the above problems in the prior art, that is, to solve the problems in the prior art of incomplete line extraction, improper handling of overlapping lines and incorrect attribute text association, which lead to low accuracy of the generated line map, the first aspect of the present invention proposes a line merging method based on an airport specification AIP map, the method comprising:
[0005] S100, fusing the airport map in the airport regulations with the parking location indication map;
[0006] S200, combining the satellite image and removing the explanatory text in the fused image to obtain a first image;
[0007] S300, performing line detection on the first image through a deep contour perception network, and merging the detected lines to obtain a line image;
[0008] The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence;
[0009] The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate a line prediction result.
[0010] In some preferred embodiments, the deformable graph encoder includes an input layer, multiple deformable convolutional layers, a batch normalization layer, an activation function layer, and an output layer connected in sequence; except for the first one of the multiple deformable convolutional layers, the input of other deformable convolutional layers is a feature graph of the input and output residual connection of the previous deformable convolutional layer.
[0011] In some preferred embodiments, the multi-head attention feature aggregator includes an input layer, a multi-head attention layer, a LayerNorm layer, a feedforward neural network layer, and an output layer connected in sequence; the input of the LayerNorm layer is a feature map of the input and output residual connection of the multi-head attention layer.
[0012] In some preferred embodiments, the pyramid feature generator includes an input layer, a pyramid pooling module, a convolution layer, a feature fusion layer, and an output layer connected in sequence; the pyramid pooling module is used to divide the input feature map into regions of different sizes and perform pooling operations on them respectively, upsample the pooled features and splice them with the input feature map.
[0013] In some preferred embodiments, the detected lines are merged by:
[0014] For short discontinuous line segments: the probability of line segment connection is calculated based on the distance and angle of the line segments, and the adjacent line segments are connected according to the probability;
[0015] For overlapping line segments: extract the features of the lines and construct an adjacency matrix, calculate the Laplacian matrix based on the adjacency matrix, solve the eigenvalues of the Laplacian matrix, and select the eigenvectors corresponding to the first K smallest eigenvalues for clustering;
[0016] Association between lines and attribute text: Based on the distance between the center of the text and the nearest point of the line, the association between the text and the line is calculated and associated.
[0017] In some preferred embodiments, the probability of line segment connection is calculated based on the distance and angle of the line segments, and the method is:
[0018] ;
[0019] in, represents the probability of line segment connection, , Represents two line segments to be connected. and Represent the starting points of the two line segments. is the angle between the unit vectors of the two line segments, represents the Euclidean distance between two points, , , Represents the standard deviation of distance, angle, and length respectively.
[0020] In some preferred embodiments, the clustering method used during clustering includes k-means clustering.
[0021] In some preferred implementations, the degree of association between text and lines is calculated as follows:
[0022] ;
[0023] in, Indicates the degree of association, Indicates the distance between the center of the text and the nearest point of the line. , Respectively represent the main direction of text and the main direction of lines. Indicates text, Represents a line.
[0024] The second aspect of the present invention provides a line merging system based on an airport specification AIP diagram, comprising:
[0025] An image fusion module configured to fuse the airport map in the airport regulations with the parking position indication map;
[0026] a text removal module configured to remove the explanatory text in the fused image in combination with the satellite image to obtain a first image;
[0027] A line merging module, configured to perform line detection on the first image through a deep contour perception network, merge the detected lines, and obtain a line image;
[0028] The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence;
[0029] The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate a line prediction result.
[0030] Beneficial effects of the present invention:
[0031] The present invention solves the problems of incomplete line extraction, improper handling of overlapping lines and incorrect attribute text association in the prior art, improves the accuracy and efficiency of line extraction and merging, and improves the accuracy of generated line drawings, reduces the need for manual intervention, accelerates the digitization process of airport layout, and provides more accurate and timely information support for aviation safety management and airport operations. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings.
[0033] Figure 1 The present invention is a flowchart of a line merging method based on an airport specification AIP diagram according to an embodiment of the present invention. DETAILED DESCRIPTION
[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0035] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.
[0036] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.
[0037] The line merging method based on the airport specification AIP diagram of the present invention comprises the following steps:
[0038] S100, fusing the airport map in the airport regulations with the parking location indication map;
[0039] S200, combining the satellite image and removing the explanatory text in the fused image to obtain a first image;
[0040] S300, performing edge detection and line extraction on the first image, and repairing the extracted lines to obtain a second image;
[0041] S300, performing line detection on the first image through a deep contour perception network, and merging the detected lines to obtain a line image;
[0042] The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence;
[0043] The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate a line prediction result.
[0044] In order to more clearly illustrate the line merging method based on the airport details AIP diagram of the present invention, the following is combined with the attached Figure 1 Each step in an embodiment of the method of the present invention is described in detail.
[0045] S100, fusing the airport map in the airport regulations with the parking location indication map;
[0046] In this embodiment, image fusion is performed by a deep feature guided multimodal image fusion and calibration algorithm (DF-MFCA), as follows:
[0047] The weights are calculated based on feature similarity, contrast, and gradient direction consistency, and then the airport map in the airport details and the parking location indication map are weightedly fused. As shown in the following formula:
[0048] (1)
[0049] (2)
[0050] in, represents the fused image, Indicates the airport diagram in the airport details, Indicates the stop position indication diagram, represents the weight, represents the coordinates (pixel coordinates in the image), , , Indicates the relative importance of feature similarity, contrast, and gradient direction consistency in weight calculation. Indicates The similarity of the depth features of the images is calculated, that is, the distance of the depth features in different images is calculated. In the present invention, the Euclidean distance is preferred. Represents the contrast of the image (i.e., calculating the difference in the gradient direction of different images at the same pixel coordinates, and then calculating the cosine of the difference), Indicates the consistency of gradient direction. The contrast of the image is obtained as follows:
[0051] (3)
[0052] in, represents the standard deviation in the local area, , Represents the local window radius.
[0053] S200, combining the satellite image and removing the explanatory text in the fused image to obtain a first image;
[0054] In this embodiment, the fused image is first preprocessed to obtain a preprocessed image; the preprocessing includes histogram equalization, Gaussian blur and sharpening filtering.
[0055] Perform text detection on the preprocessed image and output the text bounding box and text type; specifically:
[0056] For each bounding box, OCR is used to recognize text, and natural language processing (NLP) is used to obtain the text type, which includes descriptive text and attribute text.
[0057] A mask is set according to the text type, and mask processing is performed on the pre-processed image; specifically:
[0058] If the text type of the text in the bounding box is descriptive text, the corresponding position in the mask is set to 0, if it is attribute text, it is set to 1, and then the preprocessed image is processed using the mask.
[0059] (4)
[0060] in, represents the image after mask processing, represents the preprocessed image, Indicates a mask.
[0061] The straight lines in the masked image are detected using Hough transform, and the lines are enhanced using morphology; after enhancement, the text attributes are restored through inverse masking and image interpolation processing is performed; specifically:
[0062] (5)
[0063] (6)
[0064] in, represents the image after interpolation processing, represents image interpolation, represents the preprocessed image, Represents a mask, represents the image after morphologically enhancing lines, represents morphological operations, represents the Hough transform, represents the angular resolution, represents the distance resolution, represents the threshold value, Represents a structural element.
[0065] The interpolated image is fused with the satellite image in multi-scale to obtain a first image.
[0066] S300, performing line detection on the first image through a deep contour perception network, and merging the detected lines to obtain a line image;
[0067] Traditional line detection methods (such as Canny edge detection or Hough transform) may produce a large number of false positive or false negative results in complex backgrounds. DCPN (Deep Contour Perception Network) uses the powerful ability of deep learning. The network can understand the semantics and contextual relationship of lines. In this embodiment, the deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence;.
[0068] The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image;
[0069] The deformable graph encoder includes an input layer, multiple deformable convolution layers, a batch normalization layer, an activation function layer, and an output layer connected in sequence; except for the first one of the multiple deformable convolution layers, the input of the other deformable convolution layers is a feature map of the residual connection of the input and output of the previous deformable convolution layer; that is, the original image or the feature map extracted in the previous stage is obtained as input, and then the input is convolved with the deformable convolution kernel, and the final output of the multiple deformable convolution layers is batch normalized (or the result of the residual connection of the input and output of each deformable convolution layer can be batch normalized), and finally activation processing (preferably ReLU or Leaky ReLU) is performed to increase nonlinear characteristics.
[0070] The multi-head attention feature aggregator includes an input layer, a multi-head attention layer, a LayerNorm layer, a feedforward neural network layer, and an output layer connected in sequence; the input of the LayerNorm layer is a feature map of the input and output residual connection of the multi-head attention layer; that is, the feature map output by the deformable graph encoder is decomposed into multiple heads, and each head independently performs the self-attention mechanism. The output of each head is spliced together and then passed through a linear transformation layer; the feedforward neural network layer includes two fully connected layers, and the middle layer is usually much wider than the input layer to increase the expressive power of the model.
[0071] The pyramid feature generator includes an input layer, a pyramid pooling module, a convolution layer, a feature fusion layer, and an output layer connected in sequence; the pyramid pooling module is used to divide the input feature map into regions of different sizes and perform pooling operations on them respectively, upsample the pooled features back to the original size and splice them with the input feature map; perform convolution operations on the spliced features and perform fusion processing;
[0072] The DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; that is, different sizes of convolution kernels, such as 3×3, 5×5, and 7×7, are used for the features extracted by the feature extraction module to capture edge features of different scales, and then a small fully connected network is introduced to learn edge detection thresholds in different regions to adapt to different image contents and noise levels. For the features output by the fully connected network, one branch of the two-stream network focuses on color and texture information, and the other branch processes depth information. Finally, the features of the two streams are fused at a higher level, and the contours are refined and denoised through a series of convolution and deconvolution operations for the fused features.
[0073] The MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; that is, the MSF fusion module first extracts multi-scale features from the input features through a multi-scale feature extractor, and then dynamically calculates the weight of each scale feature according to the current input feature map, and then adopts a weighted summation method to fuse the features of different scales according to the dynamically calculated weights. In addition, the MSF fusion module can also add a self-attention guided fusion layer at the end, which calculates the weights of different positions of the fused feature map through the self-attention mechanism, so that the network learns to focus on which areas of features should be fused first.
[0074] The line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate line prediction results. That is, through feature maps of different levels, Hough transform or similar algorithms are used to predict lines, and a probability layer is used to assign a probability value to each predicted line to reflect the credibility of its existence.
[0075] For the lines detected by the deep contour perception network, especially for line segment connection, overlapping line processing and attribute text association, the present invention can introduce some innovative technologies to improve the accuracy and efficiency of merging.
[0076] 1) Traditional line segment connection is often based on simple distance and angle criteria, but this method is easily affected by noise. The present invention proposes to use probability weighted line segment connection, which considers the line segment length, direction consistency and the distribution probability of adjacent line segments when connecting line segments. That is, for short and discontinuous line segments, the probability of line segment connection is calculated based on the distance and angle of the line segments, and adjacent line segments with probabilities greater than the set threshold are connected;
[0077]
[0078] in, represents the probability of line segment connection, , Represents two line segments to be connected. and Represent the starting points of the two line segments. is the angle between the unit vectors of the two line segments, represents the Euclidean distance between two points, , , represent the standard deviation of distance, angle and length respectively, Represents the absolute difference between the lengths of two line segments.
[0079] 2) Traditional overlapping line processing methods may be too simple and cannot effectively process dense or interlaced lines. The present invention combines spectral clustering algorithms and deep learning, learns the feature representation of lines through neural networks, and then uses spectral clustering algorithms to cluster these representations to solve the problem of overlapping lines. That is, for overlapping line segments: extract the features of the lines and construct an adjacency matrix, calculate the Laplacian matrix based on the adjacency matrix, solve the eigenvalues of the Laplacian matrix, and select the eigenvectors corresponding to the first K smallest eigenvalues for clustering; the details are as follows:
[0080]
[0081] represents the extracted line features, represents the adjacency matrix;
[0082] Calculate the Laplace matrix L = D − A, where D is a diagonal matrix with diagonal elements D ii =∑ j A ij Next, solve the eigenvalue problem LX=λX, take the eigenvectors corresponding to the first K smallest eigenvalues, and perform k-means clustering to obtain line clusters.
[0083] 3) Traditional attribute text association methods may only match based on distance, while ignoring the visual association between text and lines. The present invention introduces a visual attention mechanism to optimize the association by analyzing the spatial relationship and visual clues between text and lines. That is, based on the distance between the center of the text and the nearest point of the line, the association between the text and the line is calculated, and the text and lines with an association greater than a set threshold are associated and matched. The association calculation method is:
[0084]
[0085] in, Indicates the degree of association, Indicates the distance between the center of the text and the nearest point of the line. , Respectively represent the main direction of text and the main direction of lines. Indicates text, Represents a line.
[0086] After obtaining the line image, text is extracted from the line image to obtain the attribute information of the line; the line image and the attribute information of the line are combined to generate the airport line map. After generating the airport line map, the runway detection is also performed using the Hough transform based on the airport satellite image. Combined with the obtained airport line map, the runway centerline is identified and its attribute is set to Runway. At the same time, according to the civil aviation standards, the attributes of the runway centerline are given, such as color, material, etc.
[0087] Take the runway as the seed and use the seed algorithm to determine the type of taxiway (T) that intersects the runway (taxiway, fast exit), and assign its attributes (taxiway, fast taxiway, color, material, etc.). Then determine whether other lines that intersect with the T are taxiways or parking stand indicators. If it is a taxiway, assign it according to the previous taxiway attributes; if it is a parking stand indicator, the current parking stand information needs to be recorded at the same time. Establish an airport database and store it with the airport ICAO as the unique identification information. After completing the assignment of airport marking information, write it into the airport database.
[0088] A line merging system based on an airport specification AIP diagram according to a second embodiment of the present invention comprises:
[0089] An image fusion module configured to fuse the airport map in the airport regulations with the parking position indication map;
[0090] a text removal module configured to remove the explanatory text in the fused image in combination with the satellite image to obtain a first image;
[0091] A line merging module, configured to perform line detection on the first image through a deep contour perception network, merge the detected lines, and obtain a line image;
[0092] The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence;
[0093] The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, a feature fusion layer, and an attention-guided fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate a line prediction result.
[0094] It should be noted that the line merging system based on the airport regulations AIP diagram provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.
[0095] A line merging device based on an airport details AIP diagram according to the third embodiment of the present invention comprises: at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned line merging method based on the airport details AIP diagram.
[0096] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned line merging method based on the airport specification AIP diagram.
[0097] Technicians in the technical field can clearly understand that, for the convenience and simplicity of description, the specific working process and related instructions of the line merging device and computer-readable storage medium based on the airport regulations AIP diagram described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.
[0098] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.
[0099] The terms "first", "second", etc. are used to distinguish similar objects rather than to describe or indicate a particular order or sequence.
[0100] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus / device.
[0101] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.
Claims
1. A line merging method based on an airport specification AIP diagram, characterized in that: The method comprises the following steps: S100, fusing the airport map in the airport regulations with the parking location indication map; S200, combining the satellite image and removing the explanatory text in the fused image to obtain a first image; S300, performing line detection on the first image through a deep contour perception network, and merging the detected lines to obtain a line image; The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence; The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate line prediction results; The deformable graph encoder comprises an input layer, a plurality of deformable convolutional layers, a batch normalization layer, an activation function layer, and an output layer connected in sequence; except for the first deformable convolutional layer, the input of the other deformable convolutional layers is a feature graph of the residual connection of the input and output of the previous deformable convolutional layer; The multi-head attention feature aggregator includes an input layer, a multi-head attention layer, a LayerNorm layer, a feedforward neural network layer, and an output layer connected in sequence; the input of the LayerNorm layer is a feature map of the input and output residual connection of the multi-head attention layer; The pyramid feature generator includes an input layer, a pyramid pooling module, a convolution layer, a feature fusion layer, and an output layer connected in sequence; the pyramid pooling module is used to divide the input feature map into regions of different sizes and perform pooling operations on them respectively, upsample the pooled features and splice them with the input feature map.
2. The line merging method based on the airport details AIP diagram according to claim 1 is characterized in that: The detected lines are merged as follows: For short discontinuous line segments: the probability of line segment connection is calculated based on the distance and angle of the line segments, and the adjacent line segments are connected according to the probability; For overlapping line segments: extract the features of the lines and construct an adjacency matrix, calculate the Laplacian matrix based on the adjacency matrix, solve the eigenvalues of the Laplacian matrix, and select the eigenvectors corresponding to the first K smallest eigenvalues for clustering; Line and attribute text association: Based on the distance between the center of the text and the nearest point of the line, the association between the text and the line is calculated and associated.
3. The line merging method based on the airport details AIP diagram according to claim 2 is characterized in that: The probability of connecting line segments is calculated based on the distance and angle of the line segments. The method is: ; in, represents the probability of line segment connection, , Represents two line segments to be connected. and Represent the starting points of the two line segments. is the angle between the unit vectors of the two line segments, represents the Euclidean distance between two points, , , Represents the standard deviation of distance, angle, and length respectively.
4. The line merging method based on the airport details AIP diagram according to claim 3 is characterized in that: The clustering method used in clustering includes k-means clustering.
5. The line merging method based on the airport details AIP diagram according to claim 4 is characterized in that: The method for calculating the correlation between text and lines is: ; in, Indicates the degree of association, Indicates the distance between the center of the text and the nearest point of the line. , Respectively represent the main direction of text and the main direction of lines. Indicates text, Represents a line.
6. A line merging system based on the airport specification AIP diagram, characterized in that: The system includes: An image fusion module configured to fuse the airport map in the airport details with the parking position indication map; a text removal module configured to remove the explanatory text in the fused image in combination with the satellite image to obtain a first image; A line merging module, configured to perform line detection on the first image through a deep contour perception network, merge the detected lines, and obtain a line image; The deep contour perception network includes a feature extraction module, a DCM module, an MSF fusion module, and a line prediction module connected in sequence; The feature extraction module includes a deformable graph encoder, a multi-head attention feature aggregator, and a pyramid feature generator connected in sequence; the feature extraction module is used to extract multi-level features of the first image; the DCM module includes a multi-scale edge detector, a fully connected network, a two-stream network, convolution and deconvolution operations connected in sequence; one branch of the two-stream network is used to process color and texture features, and one branch is used to process depth features, and the features processed by the two branches are fused; the DCM module is used to enhance line feature learning based on the extracted multi-level features; the MSF fusion module includes a multi-scale feature extractor, a dynamic weight allocator, and a feature fusion layer connected in sequence; the MSF fusion module is used to fuse feature maps of different scales; the line prediction module includes a multi-level line detector and a probability layer connected in sequence; the line prediction module is used to generate line prediction results; The deformable graph encoder comprises an input layer, a plurality of deformable convolutional layers, a batch normalization layer, an activation function layer, and an output layer connected in sequence; except for the first deformable convolutional layer, the input of the other deformable convolutional layers is a feature graph of the residual connection of the input and output of the previous deformable convolutional layer; The multi-head attention feature aggregator includes an input layer, a multi-head attention layer, a LayerNorm layer, a feedforward neural network layer, and an output layer connected in sequence; the input of the LayerNorm layer is a feature map of the input and output residual connection of the multi-head attention layer; The pyramid feature generator includes an input layer, a pyramid pooling module, a convolution layer, a feature fusion layer, and an output layer connected in sequence; the pyramid pooling module is used to divide the input feature map into regions of different sizes and perform pooling operations on them respectively, upsample the pooled features and splice them with the input feature map.
