A method for urban waterlogging risk assessment based on target recognition of satellite remote sensing images

Abstract

The present invention provides a method for urban waterlogging risk assessment based on target recognition of satellite remote sensing images, which is characterized by including: an urban waterlogging point collection module: obtaining urban waterlogging point data through a social media platform, obtaining the longitude and latitude information of the waterlogging points by using ArcGIS software, and respectively obtaining the corresponding satellite remote sensing images and elevation data from the TianDiWang and the Geospatial Data Cloud Platform according to the longitude and latitude information; a satellite remote sensing image feature extraction module: inputting the satellite remote sensing image into a deep learning model, identifying the target classes in the satellite remote sensing image, and taking the sum of the pixel numbers of each identified target as the feature value of the urban waterlogging influencing factor; an elevation data extraction module: downloading through the Geospatial Data Cloud Platform to obtain elevation tif data centered on the waterlogging points, and then extracting the elevation value and the relative elevation value of the waterlogging points, including six features such as water bodies, roads, green spaces, elevation, etc.; a prediction and analysis module based on the XGBoost model: integrating the obtained feature values and elevation values into a data set, training the XGBoost model, and analyzing the influencing factors of urban waterlogging risk through each index weight.

Description

Technical Field

[0001] The present invention relates to a technology for analyzing factors affecting urban waterlogging disasters, and more specifically, to a method for studying urban waterlogging disaster risks and identifying and analyzing influencing factors. Background Art

[0002] Urban waterlogging disasters have a huge impact on cities, causing immeasurable losses to cities: waterlogging in urban areas, traffic paralysis, threats to houses and property, pollution of water resources, rapid spread of diseases, destruction of buildings and facilities, and even possible casualties; flash floods in remote urban areas and poor drainage; in addition to waterlogging in urban areas, surrounding farmland and villages will be flooded, crop yields will be severely reduced, livestock and people’s lives and property will be threatened, and serious economic losses will be caused.

[0003] The research methods of urban waterlogging disaster risk usually use historical disaster data, scenario analysis, remote sensing and GIS, indicator system, etc. to conduct risk research. The evaluation process based on historical disaster data is relatively simple, but it is highly dependent on historical data, so it is necessary to continuously introduce new disaster data to ensure the reliability of the research results. The risk assessment method based on scenario analysis improves the spatial accuracy of risk assessment results, but has high requirements for the time scale, accuracy, and simulation modeling of the data. The risk assessment method based on the combination of remote sensing technology and GIS also has problems such as large research scale and low mapping accuracy. The evaluation method based on the indicator system is a commonly used method in urban flood risk assessment. The flood disaster risk index is used to construct a regional risk assessment model. Summary of the invention

[0004] The present invention provides an urban waterlogging risk assessment method based on satellite remote sensing image target recognition, which is used to solve the problem that the urban waterlogging disaster risk research method is incomplete.

[0005] A method for urban waterlogging risk assessment based on satellite remote sensing image target recognition, comprising:

[0006] Urban waterlogging point collection module: obtain urban waterlogging point data through social media platforms, use ArcGIS software to obtain the longitude and latitude information of waterlogging points, and obtain corresponding satellite remote sensing images and elevation data from Tiandiwang and the geospatial data cloud platform based on the longitude and latitude information;

[0007] Satellite remote sensing image feature extraction module: input satellite remote sensing images into the deep learning model, identify the target class in the satellite remote sensing image, and use the sum of the number of pixels of each identified target as the feature value of the influencing factor of urban waterlogging;

[0008] Elevation data extraction module: Download through the Geospatial Data Cloud Platform to obtain elevation tif data centered on the waterlogging points, and then extract the elevation values and relative elevation values of the waterlogging points;

[0009] Prediction and analysis module based on the XGBoost model: Integrate the obtained feature values and elevation values into a data set, train the XGBoost model, and analyze the influencing factors of urban waterlogging risk through the weights of each index.

[0010] As an embodiment of the present invention, the urban waterlogging point collection module includes:

[0011] Waterlogging point acquisition unit: By crawling web pages containing keywords "drowning / flooding" or "waterlogging / water accumulation" in news reports from 2017 to 2018, clean the obtained text data, delete duplicates and information irrelevant to flood disasters, and then preprocess the text through Chinese word segmentation and stop word removal to obtain more than 70,000 pieces of information. To geolocate urban waterlogging points from the text content, the community directories of some cities in China across the country were also downloaded, including information such as community names and geographical locations, which were obtained from the well-known housing website https: / / www.anjuke.com / . Terms related to communities, roads, and directions were extracted from the posts. Subsequently, the community directories of some cities in China were used to match these terms, so as to determine the geographical locations of the reported waterlogging points. Import these geographical locations into ArcGIS software to obtain the corresponding longitude and latitude coordinates, which is convenient for obtaining satellite images of the waterlogging points;

[0012] Satellite remote sensing image acquisition unit: By using the geocoding and reverse geocoding of the National Geospatial Information Public Service Platform Tianditu, intercept a satellite remote sensing image with a resolution of 1024×1024 on Tianditu with the waterlogging point coordinates as the central area.

[0013] As an embodiment of the present invention, the satellite remote sensing image feature extraction module includes:

[0014] Annotation data unit: Use the labelme tool to annotate water bodies, roads, and green space target classes in satellite remote sensing images, and use the annotated data as the training set of the deep learning model. A total of 200 satellite remote sensing images and 4 target classes were annotated;

[0015] Deep learning training unit: Under the Pytorch framework, use the training set to train the semantic segmentation U-net model, and verify and optimize it through the test set. Finally, the mean pixel accuracy (MPA) reached about 81.89%.

[0016] As an embodiment of the present invention, the elevation data extraction module includes:

[0017] Elevation value extraction unit: Extract the elevation data of the center point of the elevation tif file matrix through a program, that is, obtain the elevation value of the waterlogging point;

[0018] Relative elevation value extraction unit: Since there are differences in the overall elevation between different cities, in order to make the elevation data between cities meaningful for joint research, the average elevation of the four vertices is subtracted from the elevation of the center point of the elevation tif image as the relative elevation of the sample point.

[0019] As an embodiment of the present invention, the prediction analysis module based on the XGBoost model includes:

[0020] XGBoost method unit: Provide the basis for constructing a training model through the XGBoost method of a preset machine learning system, verify and optimize the XGBoost model using a test set, evaluate the performance of the model, and finally obtain the weights of the urban waterlogging risk factors;

[0021] Model optimization unit: Use the grid tuning method to determine the optimal parameters of the model, and finally verify the model using a test set. The average value of the final AUC reaches about 0.88 through 5-fold cross-validation.

[0022] The beneficial effects of the present invention are as follows: Apply satellite remote sensing data and deep learning to study the risk of urban waterlogging disasters. Construct a method model based on deep learning, extract the features of water bodies, roads, and green spaces in satellite remote sensing images, and finally analyze the main image factors of the risk of urban waterlogging disasters through the XGBoost model, providing more research methods for urban waterlogging risk factors and improving the research efficiency.

[0023] The technical solution of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0024] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the accompanying drawings:

[0025] Figure 1 It is a flowchart of a method for evaluating the risk of urban waterlogging based on satellite remote sensing image target recognition in an embodiment of the present invention;

[0026] Figure 2 It is an effect diagram of satellite remote sensing image feature extraction in a method for evaluating the risk of urban waterlogging based on satellite remote sensing image target recognition in an embodiment of the present invention. Detailed Embodiments

[0027] The preferred embodiments of the present invention will be described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0028] An embodiment of the present invention provides a method for urban waterlogging risk assessment based on satellite remote sensing image target recognition. The overall framework is as Figure 1 shown:

[0029] The present invention collects reports on urban waterlogging from 2017 to 2018 through social media platforms. After locating the waterlogging points, the corresponding satellite remote sensing images and elevations are obtained. Then, a training data set is made to train the semantic segmentation model and evaluate the performance of the model. The trained weights are used to predict the satellite remote sensing images of the obtained positive and negative samples. After prediction, each picture will obtain a corresponding label map (such as Figure 2 ). Next, the number of each type of pixel point on the label map is accumulated through a program, that is, the sum of the number of each type of pixel point obtained, that is, the area of each category. Secondly, the area of each category in each picture, the elevation value of the corresponding picture, the relative elevation value, and the labels ("1" or "0") of the corresponding positive and negative samples are integrated together and saved in the format of a comma-separated value (CSV) file to form the final data table. Finally, the data table is sent into the Extreme Gradient Boosting model to analyze the urban waterlogging disaster factors and evaluate the performance of the model.

[0030] An embodiment of the present invention provides a method for urban waterlogging risk assessment based on satellite remote sensing image target recognition. The present invention includes:

[0031] Urban waterlogging point acquisition module: Obtain urban waterlogging point data through social media platforms, use ArcGIS software to obtain the longitude and latitude information of the waterlogging points, and obtain the corresponding satellite remote sensing images and elevation data from the TianDiWang and Geospatial Data Cloud platforms respectively according to the longitude and latitude information;

[0032] Satellite remote sensing image feature extraction module: Input the satellite remote sensing image into a deep learning model, identify the target classes in the satellite remote sensing image, and use the sum of the pixel point numbers of each identified target as the feature value of the urban waterlogging influencing factor;

[0033] Elevation data extraction module: Download through the Geospatial Data Cloud platform to obtain elevation tif data centered on the waterlogging point, and then extract the elevation value and relative elevation value of the waterlogging point;

[0034] Prediction and analysis module based on the XGBoost model: Integrate the obtained feature values and elevation values into a data set, train the XGBoost model, and analyze the influencing factors of urban waterlogging risk through each index weight.

[0035] The working principle of the above technical solution is as follows: The present invention divides the obtained satellite images of urban waterlogging into a training set and a test set, and constructs a training sample library; under the Pytorch framework, the sample data of the training sample library is trained through a U-Net network to generate a semantic segmentation model; the result obtained after training the semantic segmentation model is then integrated with the elevation data of the inundation points and imported into the XGBoost model to obtain information on the main influencing factors of urban waterlogging disasters.

[0036] The beneficial effects of the above technical solution are as follows: By applying satellite image data related to urban waterlogging disasters and deep learning to analyze the influencing factors of flood disasters. A method model based on deep learning is constructed to extract features of water bodies, roads, and green space elements in satellite images, and finally, the main influencing factors of urban waterlogging are obtained. The overall recognition rate of the recognition of water bodies, roads, and green space targets in satellite images by the present invention is greatly improved (it can reach 81.89% during actual implementation), and it has a more accurate recognition ability compared with the existing technologies. It shows that satellite image pictures can be better recognized by the deep learning model used in the present invention, can effectively automatically extract water bodies, roads, and green space targets in satellite image pictures, and can accurately analyze the main influencing factors of urban flood disasters.

[0037] In one embodiment, the urban waterlogging point acquisition module includes:

[0038] Waterlogging point acquisition unit: By crawling information containing keywords "drowning / flooding" or "waterlogging / water accumulation" in news reports from 2017 to 2018, cleaning the obtained text data, deleting duplicates and information irrelevant to flood disasters, and then preprocessing the text through Chinese word segmentation and stop word removal, more than 70,000 pieces of information are obtained, and 6,407 waterlogging points are obtained. In order to geolocate urban waterlogging points from the text content, a community directory of some cities in China nationwide is downloaded, including information such as community names and geographical locations, and this information comes from the well-known housing website https: / / www.anjuke.com / . Terms related to communities, roads, and directions are extracted from the posts. Subsequently, the community directory of some cities in China is used to match these terms, so that the geographical locations of the reported waterlogging points can be determined. These geographical locations are imported into the ArcGIS software to obtain the corresponding longitude and latitude coordinates, which is convenient for obtaining satellite images of waterlogging points;

[0039] Satellite remote sensing image acquisition unit: By using the geocoding and reverse geocoding of the National Geographic Information Public Service Platform Tianditu, a satellite remote sensing image with a resolution of 1024×1024 is intercepted on Tianditu with the waterlogging point coordinates as the central area.

[0040] In one embodiment, the satellite remote sensing image feature extraction module includes:

[0041] Annotated data unit: The labelme tool is used to annotate water bodies, roads, and green space target classes in satellite remote sensing images. After each class is annotated, the corresponding label name is filled in. After each image is annotated and saved, a json file will be generated, which contains label information. Finally, the json file is converted into a dataset for use as the training set of the deep learning model. A total of 200 satellite remote sensing images are annotated, including 4 target classes: water bodies, roads, green spaces, and others;

[0042] Deep learning training unit: Under the Pytorch framework, the training set is used to train the semantic segmentation U-net model, and it is verified and optimized through the test set. Finally, the mean pixel accuracy (MPA) reaches about 81.89%.

[0043] In one embodiment, the elevation data extraction module includes:

[0044] Elevation value extraction unit: The elevation data of the center point of the elevation tif file matrix is extracted through a program, that is, the elevation value of the water accumulation point is obtained;

[0045] Relative elevation value extraction unit: Since there are differences in the overall elevation between different cities, in order to make the elevation data between cities meaningful for joint research, the average elevation of the four vertices is subtracted from the elevation of the center point of the elevation tif image as the relative elevation of the sample point.

[0046] In one embodiment, the prediction and analysis module based on the XGBoost model includes:

[0047] XGBoost method unit: Provide the basis for constructing the training model through the XGBoost method of the preset machine learning system, verify and optimize the XGBoost model using the test set, and evaluate the performance of the model. Finally, the weights of the urban waterlogging risk factors are obtained;

[0048] Model optimization unit: Use the grid tuning method to determine the optimal parameters of the model. Finally, verify the model using the test set, and the average value of AUC reaches about 0.88 through 5-fold cross-validation.

Claims

1. A method for urban waterlogging risk assessment based on target recognition of satellite remote sensing images, characterized in that, it includes: Urban waterlogging point collection module: Obtain urban waterlogging point data through social media platforms, use ArcGIS software to obtain the longitude and latitude information of the waterlogging points, and obtain the corresponding satellite remote sensing images and elevation data from the TianDiWang and Geospatial Data Cloud platforms respectively according to the longitude and latitude information; Satellite remote sensing image feature extraction module: Input the satellite remote sensing image into a deep learning model, identify the target classes in the satellite remote sensing image, and take the sum of the pixel points of each identified target as the feature value of the urban waterlogging influencing factor; Elevation data extraction module: Download through the Geospatial Data Cloud platform to obtain elevation tif data centered on the waterlogging point, and then extract the elevation value and relative elevation value of the waterlogging point; Prediction and analysis module based on the XGBoost model: Integrate the obtained feature values and elevation values into a data set, train the XGBoost model, and analyze the influencing factors of urban waterlogging risk through each index weight; The satellite remote sensing image feature extraction module includes: Labeled data unit: Use the labelme tool to label the water body, road, and green space target classes in the satellite remote sensing image, and use the labeled data as the training set of the deep learning model; Deep learning training unit: Based on the Pytorch deep learning framework, use the training set to train the semantic segmentation U-net model, and verify and optimize it through the test set; The elevation data extraction module includes: Elevation value extraction unit: Extract the elevation data of the center point of the elevation tif file matrix through the program, that is, obtain the elevation value of the waterlogging point; Relative elevation value extraction unit: Since there are differences in the overall elevation between different cities, in order to make the elevation data between cities meaningful for joint research, use the elevation of the center point of the elevation tif image minus the average elevation of the four vertex elevations as the relative elevation of the sample point.

2. A method for urban waterlogging risk assessment based on target recognition of satellite remote sensing images according to claim 1, characterized in that, The urban waterlogging point collection module includes: Waterlogging point acquisition unit: Collect information related to urban waterlogging from 2017 to 2018 through news reports, perform deduplication, and use ArcGIS software to obtain the longitude and latitude information of the waterlogging points; Satellite remote sensing image acquisition unit: Through the geocoding and reverse geocoding of the National Geographic Information Public Service Platform TianDiTu, intercept a satellite remote sensing image with a resolution of 1024×1024 on TianDiTu with the waterlogging point coordinates as the central area.

3. A method for urban waterlogging risk assessment based on target recognition of satellite remote sensing images according to claim 1, characterized in that, The prediction and analysis module based on the XGBoost model includes: XGBoost method unit: Provide the basis for constructing a training model through the XGBoost method of a preset machine learning system; Model optimization unit: Use the grid tuning method to determine the optimal parameters of the model, and finally verify the model using the test set.

Images