Urban pipeline simulation data construction method based on intelligent remote sensing interpretation

Abstract

The invention discloses an urban pipeline simulation data construction method based on intelligent remote sensing interpretation, and relates to the technical field of remote sensing image intelligent interpretation and urban geographic information basic data rapid generation. Comprising the following steps: acquiring an urban original ortho-image; carrying out preprocessing on the original orthoimage; marking the preprocessed orthoimage according to the road contour by using an image marking tool, and prefabricating a sample training set and a verification set; optimization is carried out based on coding and decoding structures in the U-Net original model, an urban pipeline simulation data construction model is established, and training of the model is completed; road line extraction is carried out by using the trained model; segmenting the road line; and assigning attributes to the segmented points and lines to complete generation of pipeline data. According to the method, the road of the remote sensing image is intelligently and automatically extracted, and the pipeline simulation data is constructed based on the extracted road data, so that the rapid generation of the urban pipe network data is realized, and the pipeline simulation data is provided under a specific condition.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent interpretation of remote sensing images and rapid generation of basic urban geographic information data, and in particular to a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation. Background Art

[0002] The construction of urban digital twins is to use digital twin technology to build a digital twin city base that corresponds to the physical city one by one and maps to each other, so as to realize the digitization and virtualization of all elements of the city and provide real-time and accurate decision support for city managers. The core of the construction of the digital twin city base is the acquisition of basic surveying and mapping data of the city. However, with the continuous construction of urban digital twins, it is found that the cost of acquiring data is getting higher and higher. However, the acquisition of simulated data of urban underground pipe networks does not require accurate data, only simulated data is sufficient.

[0003] Traditional underground pipeline network data acquisition uses traditional manual surveying and mapping methods, such as field surveys, non-destructive testing technology, underground pipeline census, and pipeline network drawing analysis. Although it can provide relatively accurate pipeline network data, it is costly and inefficient. In some specific scenarios of digital twins, the data accuracy requirements for "non-focus areas" are not high, and excessive accuracy may not be in line with the principle of economic efficiency.

[0004] Therefore, it is an urgent problem for technical personnel in this field to propose a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation to solve the difficulties existing in the prior art. Summary of the invention

[0005] In view of this, the present invention provides a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation. It can realize the rapid generation of urban pipeline network data by intelligently and automatically extracting roads from remote sensing images and constructing pipeline simulation data based on the extracted road data, and provide pipeline simulation data under specific circumstances.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation, comprising:

[0008] S1. Obtain the original orthophoto of the city;

[0009] S2, preprocessing the original orthophoto to obtain a preprocessed orthophoto;

[0010] S3, using image annotation tools to annotate the preprocessed orthophoto according to the road contour, and preform a sample training set and a validation set;

[0011] S4, based on the optimization of the encoding and decoding structure in the original U-Net model, the urban pipeline simulation data construction model is established to complete the model training;

[0012] S5. Use the trained model to extract road lines;

[0013] S6, segmenting the road line;

[0014] S7. Assign attributes to the points and lines after segmentation to complete the generation of pipeline data.

[0015] In the above method, optionally, in S1, the original orthophoto of the city is obtained, and the original orthophoto data of the city is collected using a drone.

[0016] In the above method, optionally, in S2, the original orthophoto is preprocessed as follows:

[0017] The collected original orthophoto data of the city are stitched and orthorectified, the aerial photographs taken by the drone during takeoff and landing are manually deleted, and the photos with severe exposure and focus failure are eliminated. The synthesis of the orthophoto map is completed according to the process of original image import, aerial triangulation, model reconstruction, and result input.

[0018] In the above method, optionally, in S3, the preprocessed orthophoto is annotated according to the road contour using an image annotation tool, and preformed into a sample training set and a validation set:

[0019] The target samples were extracted according to the geometric shape and spectral characteristics of the object contours, and the open source Labelme data annotation tool was used to annotate the road contour images with semantic segmentation datasets. The annotated sample data were divided into training set and validation set in a ratio of 9:1.

[0020] In the above method, optionally, in S4, the encoding and decoding structures in the original U-Net model are optimized, a model is constructed based on the urban pipeline simulation data, and the model training is completed, specifically:

[0021] Based on the encoding and decoding structure in the original U-Net model, the optimization is performed: the original U-Net network is introduced into the Wavelet Transform module; a channel attention module is constructed and integrated into the jump connection of the original network; the Res Net residual module is introduced to obtain the optimized model network structure, and a model for building urban pipeline simulation data is established;

[0022] During the model training process: the annotated orthophoto data is divided according to the Pascal VOC data format. The original size of the image is 512 pixels * 512 pixels. According to the optimized model network structure, the number of training Epochs is set to 100, and the number of samples captured at a single time Batch size is set to 4 to complete the model training.

[0023] In the above method, optionally, in S5, a trained model is used to extract road lines: the image data of the area to be identified is input into the trained urban pipeline simulation data to build a model, the recognition result of the road network is extracted, the recognized raster result is converted into .shp format, and the center line of the road network is extracted according to the recognition result.

[0024] In the above method, optionally, in S6, the road line is segmented, and the specific method is as follows:

[0025] Pipeline data is simulated based on road data, and long line segments are spatially decomposed into multiple small line segments with a length of 0-70 meters. Various business attributes are assigned to them to form pipeline segments. The end points of each line segment are extracted separately to generate pipeline points, and geometric line and geometric point data that meet pipeline standard specifications are obtained.

[0026] In the above method, optionally, in S7, attributes are assigned to the segmented points and lines to complete the generation of pipeline data, which specifically includes:

[0027] Assign attributes to the obtained pipe segment and pipe point geometry data to complete the generation of pipeline data;

[0028] The pipe segment contains five basic attributes: pipeline unique number, starting pipe point number, ending pipe point number, starting point burial depth, and ending point burial depth;

[0029] The pipe point includes five attributes: pipe point unique number, pipe point elevation, attachments, pipe point features, and manhole cover shape;

[0030] Among the above attributes, the unique point number attribute value is specified in sequence as 0, 1, 2, 3, ... during the generation process. The starting point burial depth, the ending point burial depth and the pipe point elevation are randomly selected from 0 to 1 meter. Other attribute values ​​are randomly specified or fixed through a value list. Finally, a point table and a line table containing attributes are obtained.

[0031] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation, which has the following beneficial effects: 1) The present invention reduces data acquisition costs and improves work efficiency, and can automatically extract road line data from remote sensing images without a large amount of manual participation. It can quickly and accurately generate pipeline data, significantly reducing manpower and material resources and speeding up data acquisition; 2) The generated data is consistent with the format, geometry and attributes of the real data without distortion; 3) By quickly constructing urban pipeline layout scenarios, it can help planning and design personnel simulate the operation of urban pipeline systems, including water supply, drainage, gas, electricity and other pipeline systems; 4) Through simulation analysis, the operating efficiency, safety and reliability of the pipeline system can be evaluated, providing a scientific basis for urban planning and design. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0033] Figure 1 A flowchart of a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation provided by the present invention;

[0034] Figure 2 The pre-processed city orthophoto provided by the present invention;

[0035] Figure 3 The road surface image after road line extraction provided by the present invention;

[0036] Figure 4 The road vector line image extracted and provided by the present invention;

[0037] Figure 5 The three-dimensional modeling effect of the pipeline data provided by the present invention in 3DMax

[0038] Figure 6 The application effect of the pipeline data provided by the present invention in actual scenarios

[0039] Figure 7 The pipeline data provided by the present invention is detailed Micro features Effect;

[0040] Figure 8 It is a schematic diagram of the method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of 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.

[0042] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0043] Reference Figure 1 As shown, the present invention discloses a method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation, comprising:

[0044] S1. Obtain the original orthophoto of the city;

[0045] S2, preprocessing the original orthophoto to obtain a preprocessed orthophoto;

[0046] S3, using image annotation tools to annotate the preprocessed orthophoto according to the road contour, and preform a sample training set and a validation set;

[0047] S4, based on the optimization of the encoding and decoding structure in the original U-Net model, the urban pipeline simulation data construction model is established to complete the model training;

[0048] S5. Use the trained model to extract road lines;

[0049] S6, segmenting the road line;

[0050] S7. Assign attributes to the points and lines after segmentation to complete the generation of pipeline data.

[0051] Furthermore, in S1, the original orthophoto of the city is obtained, and the original orthophoto data of the city is collected using a drone.

[0052] Further, see Figure 2As shown, in S2, the original orthophoto is preprocessed, specifically:

[0053] Since remote sensing images may be affected by factors such as wind and light, which may cause optical distortion in aerial photos, the original collected data needs to be stitched and orthorectified, and aerial photos taken by the drone during takeoff and landing need to be manually deleted. Photos with severe exposure and focus failure also need to be removed. The synthesis of the orthophoto map is completed according to the process of original image import, aerial triangulation, model reconstruction, and result input.

[0054] Furthermore, in S3, the preprocessed orthophotos are annotated according to the road contours using image annotation tools, and preformed into sample training sets and validation sets:

[0055] Target samples are extracted according to the geometric shape of the ground feature contours, spectral characteristics, etc., and the open source Labelme data annotation tool is used to annotate the semantic segmentation dataset of the road images;

[0056] Furthermore, in order to ensure the accuracy and generalization ability of model training, the annotated sample data is divided into training set and validation set in a ratio of 9:1.

[0057] Furthermore, in S4, the encoding and decoding structures in the original U-Net model are optimized, and a model is constructed based on the urban pipeline simulation data to complete the model training, specifically:

[0058] Optimization is performed based on the encoding and decoding structure in the original U-Net model: the original U-Net network is introduced into the wavelet transform module Wavelet Transform, which can increase the information content (edges, textures) of remote sensing images and obtain higher quality images; a channel attention module is constructed and integrated into the jump connection of the original network, which can help the network better distinguish between targets and backgrounds, making the recognition results more accurate; the Res Net residual module is introduced to improve the accuracy and stability of image segmentation, better cope with complex image detection tasks, improve the generalization ability and performance of the model, and significantly improve the efficiency of road network extraction. The optimized model network structure is obtained, and a model for building urban pipeline simulation data is established;

[0059] During the model training process: the annotated orthophoto data is divided according to the PascalVOC data format. The original size of the image is 512 pixels * 512 pixels. According to the optimized model network structure, the number of training Epochs is set to 100, and the number of samples captured at a single time Batch size is set to 4 to complete the model training.

[0060] Further, see Figure 3As shown, in S5, the trained model is used to extract road lines: the image data of the area to be identified is input into the trained urban pipeline simulation data to build a model, the recognition result of the road network is extracted, the recognized raster result is converted into .shp format, and the center line of the road network is extracted according to the recognition result;

[0061] Further, refer to Figure 4 As shown, the image data is analyzed and processed through the trained graph algorithm model weight file to extract the recognition results of the road network;

[0062] The recognition results are usually expressed in raster form, that is, the image data is divided into a series of raster cells, each of which represents a specific area. In road network recognition, the raster cells can be marked as road network areas or non-road network areas. The results need to be expressed in .shp format according to different raster cells, that is, the raster data is converted into vector data. This process is achieved by converting raster cells into surface features.

[0063] In order to construct the center line of the road network, the method of graphics processing and road topology relationship is used to obtain the edge lines on both sides of the surface, sample each point to connect the lines and take the midpoint, so as to obtain the center line of the road from the road network surface elements, and refine and smooth the extracted road center line to ensure that the center line element conforms to the actual road shape.

[0064] Furthermore, in S6, the road line is segmented, and the specific method is as follows:

[0065] Because most urban pipelines are built along roads, we can simulate pipeline data based on road data. The original line vector file generally contains several line segments with large spatial length values ​​(such as line segments of several kilometers). The basic idea of ​​simulating pipeline generation is to decompose the long line segments spatially into multiple small line segments with a length range of 0-70 meters (refer to national standard CJJ61-2017: Technical regulations for urban underground pipeline detection), and assign various business attributes to them to form pipeline segments; at the same time, the two end points of each line segment are extracted separately to form pipeline points. At this point, the geometric line (pipeline segment) and geometric point (pipeline point) data that meet the pipeline standard specifications are obtained.

[0066] Further, see Figure 5 , Figure 6 , Figure 7 As shown, in S7, attributes are assigned to the points and lines after segmentation to complete the generation of pipeline data, which specifically includes:

[0067] Assign attributes to the obtained pipe segment and pipe point geometry data to complete the generation of pipeline data;

[0068] The pipe segment contains five basic attributes: pipeline unique number, starting pipe point number, ending pipe point number, starting point burial depth, and ending point burial depth;

[0069] The pipe point includes five attributes: pipe point unique number, pipe point elevation, attachments, pipe point features, and manhole cover shape;

[0070] Among the above attributes, the unique point number attribute value is specified in sequence as 0, 1, 2, 3, ... during the generation process. The starting point burial depth, the ending point burial depth and the pipe point elevation are randomly selected from 0 to 1 meter. Other attribute values ​​are randomly specified or fixed through a value list. Finally, a point table and a line table containing attributes are obtained.

[0071] In a specific embodiment, referring to Figure 8 As shown: A method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation, comprising:

[0072] Step 1: Data Collection:

[0073] To extract the original image data of the road, we use the orthophotos taken by DJI drones. The shooting accuracy can be controlled at the centimeter level through its low-altitude flight. In the process of using drones to collect images, in order to increase the quality of the images, it is necessary to reduce the impact of various negative factors on the collection, including environmental factors such as cloud cover and wind speed. According to the process of drone performance and safety inspection, planning aerial photography routes, adjusting aerial photography routes through field investigations, conducting aerial photography operations, and saving shooting data, the captured data are segmented to form a data set to be labeled in deep learning (original image).

[0074] Step 2: Image preprocessing:

[0075] Preprocessing of the dataset to be annotated (original image). Photos or videos captured by drones may be affected by elements such as meteorological conditions and lighting, resulting in inaccurate or unsatisfactory color rendering. Color correction technology can be used to adjust the color temperature, contrast, and saturation of the image to make the image closer to the real scene. Distortion repair includes distortion repair and lens distortion. The correction of perspective distortion is mainly based on the principle of perspective, with the aim of restoring the image to a standard perspective. The correction of lens distortion is based on the specific properties of the lens, aiming to make the straight lines in the image vertical or parallel again. Graphic conformity verification performs a series of inspections and evaluations on images to ensure that they are suitable for their intended use or application.

[0076] Step 3: Sample preparation:

[0077] The processed original image data is used to use the professional remote sensing image annotation tool LabelMe to manually annotate the road contours in the remote sensing image. The annotation rules are based on the roads to be annotated one by one. The contours of the road are drawn from the starting and ending points of the road to ensure the accuracy and completeness of the annotations. The annotated road contours are quality checked to ensure that there are no omissions or incorrect annotations; the annotated road contour data are organized into a sample set, including positive samples (including road contours) and negative samples (excluding road contours). The entire sample set is divided into a training set, a validation set, and a test set, with 80% training set, 10% validation set, and 10% training set to ensure a balanced ratio of positive and negative samples to avoid image imbalance in model training.

[0078] Step 4: Deep Learning:

[0079] 1. Optimization of network model structure:

[0080] During the training and testing of the model, it was found that the original U-Net semantic segmentation model had problems such as gradient disappearance and blurred edge features. The model's capabilities were not outstanding in terms of performance indicators such as mIOU, mPA, and F1 score, and the overall performance of the model was not very stable. Therefore, the above problems were solved in the following ways:

[0081] (1) Introducing the ResNet residual module

[0082] Since the upsampling operation of the decoder will cause the size of the feature map to expand, some detail information may be lost or the resolution may be reduced, which may affect the model's ability to capture subtle structures and edge information, thereby affecting the accuracy of the segmentation results. To solve this problem, jump connections are introduced in the decoder or more complex structures are used to maintain the richness and detail information of the features; ResNet34 is a representative network for solving the feature gradient problem. Adding ResNet34 to the decoder can solve the detail information problem. The ResNet34 residual module consists of a two-layer convolution structure (Constitutional layer). The residual module uses a pair of 3x3 convolution kernels. By introducing the ResNet residual module, the refinement and accuracy of the feature representation are improved, and the subtle features and structural information in the image can be better captured.

[0083] (2) Introducing wavelet transform layer and attention mechanism layer

[0084] Since drones are affected by lighting conditions, shooting angles, flight attitudes or speeds during the collection of original road images, problems such as missing local features of the collected images may occur, causing deviations in the model's understanding of the image. As a powerful signal processing tool, wavelet transform uses wavelet functions to decompose and extract image information. Wavelet functions are waveform functions of finite length that have the unique ability to perform fine analysis of information at different scales and locations, especially for image processing of non-stationary remote sensing information. Discrete wavelet transform can be expressed as a basis function, which can be expressed as follows according to the wavelet series expansion:

[0085]

[0086] Multi-scale analysis through wavelet transform can effectively extract the multi-scale features of the image, capture the image details and overall features, and thus enhance the model's ability to understand the image.

[0087] In addition, in order to further improve the model's attention to the road area, the background information of the image is ignored. Based on the original improved model, an attention mechanism layer is introduced. Its working mechanism is to determine the input feature map size (including the number of channels (C), height (H), and width (W)), and then process the channel features of the input feature map through the convolution layer inside the module, and perform attention weighting on each channel feature to obtain the channel feature result. Through the regulation of attention, the channel features that need to be paid attention to are determined, thereby increasing the generalization ability and task performance of the model, while reducing redundant information, making the model's recognition and detection capabilities more accurate.

[0088] 2. Model training and prediction

[0089] The annotated data is divided strictly in accordance with the data format of Pascal VOC. The original size of the image is 512 pixels * 512 pixels, and the ratio of the test set, validation set, and test set is 8:1:1. According to the improved model network structure, the Epoch (number of training times) is set to 100, the Batch size (the number of samples captured at a single time) is set to 4, and the model training is completed under the above parameters. In addition, in order to better qualitatively evaluate the recognition and detection performance of the optimized model, mPA (pixel accuracy ratio of classification), mIOU (mean interaction ratio), and F1 score (harmonic mean of Recall and Precision) are used. In this experiment, mPA is used to represent the comprehensive accuracy of road network pixels, mIOU is used to represent the average value of the intersection and union ratio of road labels and recognition detection results, Recall is used to identify the proportion of correctly predicted road contour pixels, and Precision represents the proportion of predicted results to actual road pixels.

[0090] The calculation methods of the above indicators are as follows:

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] Where TP and TN represent pixel predictions respectively, T is True, indicating correct classification, false positive samples are represented by FP, missed positive samples are represented by FN, and F is False, indicating incorrect classification.

[0099] The evaluation results are shown in Table 1 below:

[0100] Table 1 U-Net model performance comparison (%)

[0101] Model mPA mIOU Recall Precision F1 U-Net (original) 82.36 83.41 84.65 84.21 84.43 U-Net (optimization) 90.60 89.64 93.30 93.91 93.60

[0102] Step 5: Generate pipe network:

[0103] Use shp to generate pipeline network data, take the interpreted raster data as the input item of Python script, save the predicted input image information (parameters) in Python script, including coordinate information, projection information, etc., convert the raster data into TIF standard format, give the input image information to the converted raster data, and convert the raster data into corresponding vector elements according to the attribute value or other characteristics of the raster pixel to generate recognition result shp data; decompose the long line segment spatially in shp data into multiple small line segments with a length of 0-70 meters (refer to national standard CJJ61-2017: Technical regulations for urban underground pipeline detection), and assign various business attributes to them to form pipeline segments; at the same time, the endpoints of each line segment are extracted separately to generate pipeline points. At this point, the geometric line (pipe segment) and geometric point (pipe point) data that meet the pipeline standard specifications are obtained, and applied in practice through the system.

[0104] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0105] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation, characterized in that: include: S1. Obtain the original orthophoto of the city; S2, preprocessing the original orthophoto to obtain a preprocessed orthophoto; S3, using image annotation tools to annotate the preprocessed orthophoto according to the road contour, and preform a sample training set and a validation set; S4, based on the optimization of the encoding and decoding structure in the original U-Net model, the urban pipeline simulation data construction model is established to complete the model training; S5. Use the trained model to extract road lines; S6, segmenting the road line; S7. Assign attributes to the points and lines after segmentation to complete the generation of pipeline data.

2. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S1, the original orthophoto of the city is obtained, and the original orthophoto data of the city is collected using a drone.

3. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S2, the original orthophoto is preprocessed as follows: The collected original orthophoto data of the city are stitched and orthorectified, the aerial photographs taken by the drone during takeoff and landing are manually deleted, and the photos with severe exposure and focus failure are eliminated. The synthesis of the orthophoto map is completed according to the process of original image import, aerial triangulation, model reconstruction, and result input.

4. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S3, the image annotation tool is used to annotate the preprocessed orthophoto according to the road contour, and preform sample training sets and validation sets: The target samples were extracted according to the geometric shape and spectral characteristics of the object contours, and the open source Labelme data annotation tool was used to annotate the road contour images with semantic segmentation datasets. The annotated sample data were divided into training set and validation set in a ratio of 9:

1.

5. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S4, based on the encoding and decoding structure in the original U-Net model, the model is built based on the urban pipeline simulation data to complete the model training. Specifically: Based on the encoding and decoding structure in the original U-Net model, the optimization is performed: the original U-Net network is introduced into the Wavelet Transform module; a channel attention module is constructed and integrated into the jump connection of the original network; the Res Net residual module is introduced to obtain the optimized model network structure, and a model for building urban pipeline simulation data is established; During the model training process: the annotated orthophoto data is divided according to the PascalVOC data format. The original size of the image is 512 pixels * 512 pixels. According to the optimized model network structure, the number of training Epochs is set to 100, and the number of samples captured at a single time Batch size is set to 4 to complete the model training.

6. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S5, the trained model is used to extract road lines: the image data of the area to be identified is input into the trained urban pipeline simulation data to build a model, the recognition result of the road network is extracted, the recognized raster result is converted into .shp format, and the center line of the road network is extracted according to the recognition result.

7. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S6, the road line is segmented. The specific method is as follows: Pipeline data is simulated based on road data, and long line segments are spatially decomposed into multiple small line segments with a length of 0-70 meters. Various business attributes are assigned to them to form pipeline segments. The end points of each line segment are extracted separately to generate pipeline points, and geometric line and geometric point data that meet pipeline standard specifications are obtained.

8. The method for constructing urban pipeline simulation data based on intelligent remote sensing interpretation according to claim 1 is characterized in that: In S7, attributes are assigned to the segmented points and lines to complete the generation of pipeline data, including: Assign attributes to the obtained pipe segment and pipe point geometry data to complete the generation of pipeline data; The pipe segment contains five basic attributes: pipeline unique number, starting pipe point number, ending pipe point number, starting point burial depth, and ending point burial depth; The pipe point includes five attributes: pipe point unique number, pipe point elevation, attachments, pipe point features, and manhole cover shape; Among the above attributes, the unique point number attribute value is specified in sequence as 0, 1, 2, 3, ... during the generation process. The starting point burial depth, the ending point burial depth and the pipe point elevation are randomly selected from 0 to 1 meter. Other attribute values ​​are randomly specified or fixed through a value list. Finally, a point table and a line table containing attributes are obtained.

Images