Electric power gallery flood risk assessment method and system based on deep learning

Abstract

The invention discloses an electric power gallery flood risk assessment method and system based on deep learning, and belongs to the field of flood disasters, and the method comprises the steps: achieving the water extraction and risk grade assessment of a radar image, obtaining the radar image from a Sentinel-1 satellite, removing the noise through a Speckle filter, and carrying out the geometric correction, radiation correction and normalization processing. Data preprocessing is carried out to ensure data quality; performing image segmentation on the preprocessed image by using a U-Net model, extracting and recovering image features through an encoder and decoder structure, and realizing high-precision water body segmentation through a deep learning model; the water body extraction precision is evaluated by calculating indexes such as a confusion matrix, a recall rate, accuracy and MIOU, the flood risk level of the power equipment is evaluated according to the extracted water body data, the gallery bridge distance and the gradient, and risk evaluation is divided into five levels. According to the method, the problem of how to accurately extract the water body and evaluate the flood risk of the electric power gallery is solved, and a scientific basis is provided for protection and management of electric power facilities.

Description

Technical Field

[0001] The present invention relates to the field of flood disasters, and particularly to a method and system for flood risk assessment of power corridors based on deep learning. Background Art

[0002] Among numerous urban infrastructures, as a core component of the power transmission system, the safe operation of power corridors is directly related to the stability of the energy supply of the entire city. However, against the backdrop of climate change, frequent extreme weather events have led to an increasing number of flood disasters, posing unprecedented challenges to power corridors. Problems such as equipment damage and power supply interruption caused by floods have seriously affected the normal operation of residents' daily lives and industrial production activities, and may trigger secondary disasters in a chain reaction, resulting in incalculable economic losses and social impacts. Therefore, conducting flood risk assessment of power corridors and enhancing the ability to prevent and mitigate disasters have important practical significance.

[0003] Traditional methods for flood risk assessment of power corridors mainly rely on historical data and empirical models. These methods may have certain reference value in some cases, but due to their inability to fully consider the variability and complexity of flood disasters, there are often problems such as inaccurate assessment results and insufficient prediction ability. With the development of technology, especially the rise of deep learning technology, new solutions have been provided for flood risk assessment of power corridors.

[0004] The emergence of deep learning technology has provided us with a brand-new solution. As a machine learning method based on artificial neural networks, deep learning can automatically extract and learn complex features in data through the training of a large amount of data, and has achieved remarkable results in fields such as image recognition and natural language processing. Applying deep learning to flood risk assessment of power corridors can make full use of its powerful data processing and pattern recognition capabilities to improve the accuracy and reliability of assessment.

[0005] Currently, the research on flood risk assessment methods based on deep learning at home and abroad is still in its infancy, and the exploration of related applications in the field of power corridors is even more limited. Therefore, developing a method for flood risk assessment of power corridors based on deep learning has important innovation and application value. Summary of the Invention

[0006] In view of the above existing problems, the purpose of the present invention is to improve the risk prediction accuracy and disaster prevention and mitigation capabilities of power corridors in flood disasters based on the U-Net model and Sentinel-1 data. This method comprehensively applies deep learning technology, remote sensing data analysis, and geographic information system (GIS), and can, to a certain extent, conduct risk assessment on power equipment, evaluate its danger level in flood disasters, so that relevant departments can give early warnings and carry out emergency repairs in a timely manner.

[0007] To solve the above technical problems, a flood risk assessment method for power corridors based on deep learning is proposed, including,

[0008] Obtaining radar image data covering the power corridor and its surrounding areas through the Sentinel-1 satellite and performing preprocessing; establishing a U-Net model to perform image segmentation on the preprocessed images and extract the water body area in the radar images; re-collecting data and inputting it into the model, and evaluating the accuracy of water body extraction by calculating indicators such as the confusion matrix, recall rate, precision, and MIOU.

[0009] As a preferred solution of the flood risk assessment method for power corridors based on deep learning described in the present invention, wherein: the preprocessing includes using a Speckle filter to remove speckle noise in the radar image, selecting an appropriate filter window size to balance noise removal and image detail retention;

[0010] Converting the image pixel values into backscattering coefficients, and performing geometric correction on the image according to geographical factors and sensor characteristics to accurately perform spatial positioning of the image;

[0011] Performing radiometric correction to eliminate the radiometric differences in the image caused by changes in sensor characteristics and observation conditions, and making the image brightness and contrast consistent;

[0012] According to the geographical scope of the power corridor, cropping the image to extract the region of interest;

[0013] Performing normalization processing on the image data and normalizing the pixel values to the range of [0, 255];

[0014] Extracting all water bodies in the image data according to visual interpretation, establishing water body mask data, and segmenting the original image and the water body mask data.

[0015] As a preferred solution of a method for flood risk assessment of power corridors based on deep learning according to the present invention, wherein: the water body mask data includes all water bodies in the image extracted by visual interpretation after preprocessing the Sentinel-1 image. The interpreter manually identifies the water body area by observing the gray scale, texture and terrain features of the image, converts the water body area into binary mask data, marks the water body area in the mask as 1, and marks the non-water body area as 0 to establish the water body mask data. The original image and the water body mask data are segmented, the image is entirely cut into image blocks of 256x256, and the segmented images are renamed. The original image corresponds one by one with the water body mask, serving as the data set for subsequent deep learning to extract water bodies.

[0016] As a preferred solution of a method for flood risk assessment of power corridors based on deep learning according to the present invention, wherein: the U-Net model includes that the U-Net model performs image segmentation through a symmetric encoder and decoder structure;

[0017] The encoder extracts local and global features of the image through a convolutional layer of 3x3 in each layer and a pooling layer of 2x2 max pooling, gradually increasing the degree of abstraction of the features;

[0018] The decoder part restores the image resolution through a deconvolution layer of 2x2 upsampling, combines skip connections to retain high-resolution features, and performs water body segmentation;

[0019] The number of channels of the feature map is converted into the required number of categories through a 1x1 convolutional layer, the classification probability of each pixel is output, and different categories of the model are distinguished through a loss function to reduce misclassification.

[0020] As a preferred solution of a method for flood risk assessment of power corridors based on deep learning according to the present invention, wherein: the loss function includes using a cross-entropy loss function and an Adam optimizer to train the model, iteratively adjusting the model parameters, and monitoring the loss value and accuracy during the training process;

[0021] When the loss value is less than the set threshold 1 and the accuracy exceeds the set standard threshold 2, save the trained model file and perform further prediction. Re-collect the predicted images and perform preprocessing, use the trained model to predict the new data of the water body area, generate a water body distribution map of the input image area, and save it as a shp vector file.

[0022] As a preferred solution of a method for flood risk assessment of power corridors based on deep learning according to the present invention, wherein: the evaluation of the accuracy of water body extraction includes calculating the accuracy of water body extraction, generating an extraction water body accuracy table, and calculating the water body area and the water body mask data generated by visual interpretation;

[0023] The accuracy of water body extraction is evaluated by calculating indicators such as the confusion matrix, recall rate, precision, and mean intersection over union (mIoU).

[0024] The recall rate Recal is:

[0025]

[0026] The precision Precision is:

[0027]

[0028] where TP is true positive, FN is false negative, and FP is false positive;

[0029] The mean intersection over union (mIoU) is:

[0030]

[0031] where the calculation formula for IOU is as follows:

[0032]

[0033] where N is the total number of classes;

[0034] When the extraction accuracy meets the requirements, the extracted water body data is combined with the location information of the power corridor bridge and the regional slope to conduct a flood risk assessment and evaluate the flood risk level of power equipment;

[0035] According to the distance and slope of the power corridor bridge from the water body, the risk level is divided into five levels, providing a scientific basis for the protection and management of power facilities.

[0036] As a preferred solution of a method for flood risk assessment of power corridors based on deep learning according to the present invention, wherein: the risk levels include, according to the distance and slope of the power corridor bridge from the water body, the risk levels are divided into five levels: high risk, medium-high risk, medium risk, medium-low risk, and low risk;

[0037] The risk level class judgment criterion is:

[0038]

[0039] where d is the distance of the power corridor bridge from the water body, and slope is the slope of the power corridor bridge from the water body.

[0040] Another object of the present invention is to provide a power corridor flood risk assessment system based on deep learning. The present invention aims to solve the problems of accuracy, timeliness and manual intervention cost in flood monitoring and risk assessment in the prior art. By using the U-Net model to efficiently identify and segment water bodies from Sentinel-1 satellite radar images, combined with the data processing module to achieve rapid denoising, correction and annotation of the images, reducing manual intervention. At the same time, the risk assessment module intelligently calculates the risk level and generates protection suggestions, thereby improving the intelligent level and response speed of the entire power corridor flood disaster monitoring and prevention.

[0041] As a preferred solution of a power corridor flood risk assessment system based on deep learning according to the present invention, it is characterized in that it includes a data processing module, a model construction module, an accuracy evaluation module, and a risk assessment module;

[0042] The data processing module includes a data set construction unit and an image annotation and adjustment unit;

[0043] Obtain radar images from the Sentinel-1 satellite, transfer the original radar images to the Speckle filter for noise removal, generate denoised images for geometric correction, locate the images, and transfer the corrected images to the radiometric correction unit. Transfer the normalized images to the image annotation and adjustment unit for manual or semi-automatic annotation of water bodies, generate annotation data, uniformly adjust the annotated images, output a training data set of a fixed size, and transfer the processed data set to the model construction module;

[0044] The model construction module includes a model selection unit, an encoder unit, a decoder unit, a skip connection unit, and an output and loss unit;

[0045] The model selection unit selects the U-Net model for model construction. The encoder unit receives the training data set from the data processing module, performs calculations of multiple convolutional layers and pooling layers, generates feature maps, and transfers the feature maps to the decoder unit to perform upsampling and convolutional operations, gradually restoring the image resolution and generating intermediate output feature maps. The skip connection unit splices the corresponding encoder feature maps and decoder feature maps at each decoding stage to enhance the details of the output. The output and loss unit generates a segmentation result, calculates the loss function for the result, and feeds it back to the model selection unit;

[0046] The accuracy evaluation module receives the segmentation result and the true label output by the model, generates a confusion matrix, calculates indicators based on the confusion matrix, and feeds the result back to the model construction module to indicate whether the model needs to be retrained or optimized;

[0047] The risk assessment module receives the water body extraction results, the distance between the electric pole position and the water body, and the slope information, calculates the risk level, generates corresponding protection suggestions according to the results of the risk assessment, and records the corresponding relationship between the risk level and the suggestions.

[0048] A computer device includes a memory and a processor. The memory stores a computer program. The processor, when executing the computer program, implements the steps of the method for flood risk assessment of power corridors based on deep learning.

[0049] A computer-readable storage medium stores a computer program. The computer program, when executed by a processor, implements the steps of the method for flood risk assessment of power corridors based on deep learning.

[0050] The beneficial effects of the present invention: The present invention studies based on the U-Net deep learning model, combines multi-source remote sensing images and geographic information data, and successfully realizes the accurate assessment of the flood risk of power corridors. The experimental results show that this method has high accuracy and robustness in water body extraction and risk identification. The research results can provide a scientific basis for disaster prevention and mitigation of power facilities and have important engineering application value. The method for flood risk assessment of power corridors based on deep learning, compared with traditional methods, can exclude the influence of mountain shadows and terrain undulations and accurately extract water bodies from high-resolution Sentinel-1 data. Secondly, this method also adds a risk assessment module for power equipment, which can classify the importance of power facilities and give higher attention to the risk assessment of important power facilities. It can comprehensively consider factors such as slope, distance, and flood risk, calculate a comprehensive risk index, and more comprehensively assess the risk of power corridors. Description of the Drawings

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings, where:

[0052] Figure 1 It is the overall flowchart of a method for flood risk assessment of power corridors based on deep learning provided by an embodiment of the present invention.

[0053] Figure 2 It is the U-Net model framework of a method for flood risk assessment of power corridors based on deep learning provided by an embodiment of the present invention.

[0054] Figure 3The accuracy evaluation and risk assessment method framework of a flood risk assessment method for power corridors based on deep learning provided by an embodiment of the present invention.

[0055] Figure 4 The deep learning dataset of a flood risk assessment method for power corridors based on deep learning provided by an embodiment of the present invention.

[0056] Figure 5 Print the loss value during the training process of a flood risk assessment method for power corridors based on deep learning provided by an embodiment of the present invention.

[0057] Figure 6 Predict the water body result of a flood risk assessment method for power corridors based on deep learning provided by an embodiment of the present invention.

[0058] Figure 7 The risk assessment result of a flood risk assessment method for power corridors based on deep learning provided by an embodiment of the present invention.

[0059] Figure 8 The module schematic diagram of a flood risk assessment system for power corridors based on deep learning provided by an embodiment of the present invention. Detailed implementation manners

[0060] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0062] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments alone or selectively.

[0063] The present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention, for the convenience of explanation, the cross-sectional views showing the device structure will be locally enlarged out of the general scale, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0064] Meanwhile, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner, and outer" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be construed as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0065] Unless otherwise clearly defined and limited in the present invention, the terms "mounted, connected, and coupled" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, and may also be indirectly connected through an intermediate medium, or may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] Example 1, referring to Figures 1 - 3 , which is the first embodiment of the present invention. This embodiment provides a method for evaluating the flood risk of a power corridor based on deep learning, including:

[0067] S1: Obtain radar image data covering the power corridor and its surrounding areas through the Sentinel-1 satellite and perform preprocessing.

[0068] Specifically, the model construction and training part included in the present invention mainly includes the construction of a data set, the construction of a deep learning model, inputting the data set into the model for training, and applying the trained model to actual images to provide a real-time water body vector area for subsequent risk assessment. The specific steps are as follows:

[0069] S101 Data set construction

[0070] The deep learning method used in the present invention needs to first make a data set that meets the actual needs, and obtain radar image data covering the power corridor and its surrounding areas from Sentinel-1, that is, the Sentinel-1 satellite. Next, the establishment process of this data set will be described in detail.

[0071] S102 Data preprocessing

[0072] To ensure that the deep learning model based on Sentinel-1 data can accurately evaluate the flood risk of power corridors, a series of preprocessing operations need to be performed on Sentinel-1.

[0073] Use the Speckle filter to remove the speckle noise in the radar image and enhance the image quality:

[0074] I f (x,y) = ω * I Lee (x,y) + (1 - ω) * I Gamma (x,y)

[0075] where I f (x,y) is the filtered image, I Lee (x,y) and I Gamma (x,y) are the results of Lee filtering and Gamma-MAP filtering respectively, and ω is the adaptive weight, which is determined by the statistical characteristics of the local area:

[0076]

[0078] where σ n 2 is the noise variance, and σ l 2 is the variance of the local area.

[0079] Select an appropriate filtering window size to balance noise removal and image detail preservation, and the image pixel values need to be converted to backscattering coefficients, expressed in the standardized unit dB for subsequent processing and analysis. The backscattering coefficient conversion formula is:

[0080]

[0081] where DN is the original digital value, A k is the calibration factor, and θ is the incident angle.

[0082] According to geographical factors and sensor characteristics, perform geometric correction on the image to ensure accurate spatial positioning of the image:

[0083] x' = a0 + a1x + a2y + a3xy + a4x 2 + a5y 2 + a6x 2 y + a7xy 2 + a8x 3 + a9y 3

[0084] y' = b0 + b1x + b2y + b3xy + b4x 2 + b5y 2 + b6x2 y + b7xy 2 + b8x 3 + b9y 3

[0085] where (x', y') are the corrected coordinates, (x, y) are the original coordinates, a i and b i are undetermined coefficients.

[0086] Perform radiometric correction to eliminate the radiometric differences in the image caused by sensor characteristics and observation condition changes, and ensure the consistency of image brightness and contrast:

[0087]

[0088] where DN corrected is the corrected digital value, DN min and DN max are the minimum and maximum digital values of the image respectively, L max and L min are the ideal radiation ranges.

[0089] According to the geographical range of the power corridor, crop the image to extract the region of interest ROI. Finally, perform normalization processing on the image data to standardize the pixel values to the range of [0, 255] for the input of the deep learning model:

[0090]

[0091] where I norm(x,y) is the normalized image, I (x,y) is the original image, I min and I max are the global minimum and maximum values of the image respectively, σ l(x,y) is the local standard deviation, σ global is the global standard deviation, and α is the contrast enhancement factor.

[0092] S103 Image annotation and adjustment

[0093] The present invention extracts water bodies from Sentinel-1 images based on deep learning. In order to adapt to the geographical location of the target area and the characteristics of water body extraction, it is necessary to construct its own dataset. Select Sentinel images in different regions, at different times, and with different climates, and perform water body annotation on them. After all the water bodies in the target area are annotated, in order to improve the training speed and efficiency and make the images more stable, the training images are uniformly adjusted to a fixed size of 256x256 to improve the performance and reliability of the deep learning model.

[0094] After preprocessing the Sentinel-1 images by radiometric correction, geometric correction, noise removal, and image cropping, all water bodies in these images were extracted based on visual interpretation;

[0095] Interpreters manually identified water areas by observing the grayscale, texture, and terrain features of the image;

[0096] These water areas are then converted into binary mask data, in which the water areas are marked as 1 and the non-water areas are marked as 0. Water mask data is established as the basis for subsequent model training.

[0097] Finally, the original image and water body mask data are segmented, all images are cut into 256x256 image blocks, and the segmented images are renamed. The original image and the water body mask need to correspond one by one as the data set for the subsequent deep learning method to extract water bodies.

[0098] S2: Establish a U-Net model to perform image segmentation on the preprocessed image and extract the water area in the radar image.

[0099] Furthermore, the present invention uses the U-Net model to extract features from the Sentinel-1 radar remote sensing data. Figure 2 , identifying water bodies in radar images. U-Net is a convolutional neural network for image segmentation, especially suitable for tasks that require high-precision and high-resolution segmentation, such as medical image processing, satellite image analysis, etc.

[0100] S201 Model Architecture

[0101] The structure of the U-Net model is as follows Figure 3 As shown in the figure, the structure of the U-Net model can be divided into two symmetrical parts: the encoder-contraction path-and the decoder-expansion path; its overall structure presents a U shape, hence the name U-Net. The encoder and decoder are connected to each other through jump connections, which play a vital role in image segmentation tasks.

[0102] S202 encoder-contraction path, the encoder part is mainly responsible for extracting the features of the image:

[0103] Convolutional layers: Each stage consists of two 3x3 convolutional layers, each followed by a ReLU activation function. Convolutional layers are used to extract local features of the image.

[0104] Pooling Layer: Each stage is followed by a 2x2 max pooling layer for downsampling, halving the feature map size and increasing the depth of the feature map. This helps capture the global features of the image.

[0105] As the network goes deeper layer by layer, the spatial resolution of the feature map gradually decreases, but the number of features (number of channels) gradually increases. This enables the network to capture more abstract and advanced features.

[0106] S203 Decoder - Extended Path: The decoder part is responsible for gradually restoring the features extracted by the encoder to the resolution of the original image for accurate pixel-level segmentation:

[0107] Upsampling layer: Each stage consists of a 2x2 deconvolution layer for upsampling, doubling the size of the feature map. The deconvolution layer restores the feature map to a higher spatial resolution.

[0108] Convolutional layer: Each upsampling stage is followed by two 3x3 convolutional layers, and each convolutional layer is followed by a ReLU activation function. The convolutional layer is used to fuse the upsampled feature map with the feature map of the encoder part to restore the detail information of the original image.

[0109] S204 Jump Connection

[0110] A key feature of the U-Net model is the skip connection, which directly connects the feature map of each stage in the encoder with the feature map of the corresponding stage in the decoder. These skip connections can help the decoder partially recover the lost spatial information, thereby improving the accuracy of segmentation. The specific implementation method is to splice the feature map of the encoder with the upsampled feature map of the decoder.

[0111] S205 final output layer

[0112] In the last layer of the decoder, a 1x1 convolutional layer is used to convert the number of channels of the feature map into the required number of categories. The output of this convolutional layer is the final segmentation result, which indicates the probability that each pixel belongs to a certain category.

[0113] S206 loss function

[0114] The processed data set is input into the established U-Net model, and the cross entropy loss function and Adam optimizer are used to train the model, and the model parameters are adjusted through iteration.

[0115] Monitor the loss value and accuracy during the training process, adjust the model parameters to prevent overfitting, and when the loss value is lower than the set threshold and the accuracy is higher than the predetermined standard, save the trained model file and prepare for the next prediction;

[0116] New images that need to be predicted are collected and preprocessed. The trained model is used to predict new data in the water area, and a water body distribution map of the input image area is generated and saved as a shp vector file.

[0117] Among them, the training of the U-Net model usually uses loss functions such as cross-entropy loss function or Dice loss function. These loss functions are used to measure the difference between the predicted segmentation result and the true label, and guide the update of the model's parameters.

[0118] The U-Net model has a symmetric encoder and decoder structure, combined with skip connections, enabling it to capture both global and local information of the image while maintaining high-resolution features. This design makes the U-Net perform excellently in tasks that require fine segmentation. The success of the U-Net provides an efficient and accurate solution for image segmentation tasks.

[0119] S3: Re-collect the data and input it into the model, and evaluate the accuracy of water body extraction by calculating metrics such as the confusion matrix, recall rate, precision, and MIOU.

[0120] Furthermore, according to the method of extracting water bodies from Sentinel-1 based on deep learning, the present invention needs to apply the extracted water bodies to actual needs. Before that, we need to first judge the accuracy of the extracted water bodies. When the accuracy of the extracted water bodies meets the requirements, according to the distance between the power bridge and the water body and the specific slope of the area, evaluate its risk level of pole collapse. The framework of the accuracy evaluation and risk assessment method is attached Figure 3 。

[0121] S301 Accuracy evaluation

[0122] The deep learning water body extraction designed by the present invention is a binary classification problem. Calculate the accuracy of water body extraction and generate an accuracy table for the extracted water bodies. Calculate the generated water body area with the water body mask data generated by visual interpretation. The water body mask data generated by visual interpretation is used as the reference image, and judge the water body extraction accuracy by calculating the confusion matrix, recall rate, precision, and mean intersection over union between the result image and the visual interpretation standard image.

[0123] The confusion matrix contains four main metrics: true positive TP, true negative TN, false positive, and false negative;

[0124] Recall rate:

[0125]

[0126] Among them, TP is the number of true positives and FN is the number of false negatives.

[0127] Precision:

[0128]

[0129] Among them, TP is the number of true positives and FP is the number of false positives.

[0130] Mean intersection over union MIOU:

[0131]

[0132] Among them, TP i 、FP i 、FN i are the true positive, false positive, and false negative of the i-th class, respectively.

[0133] MIOU is the average value of the IOU of all classes:

[0134]

[0135] Among them, N is the total number of classes.

[0136] S302 Risk Assessment

[0137] Combining slope data, utility pole location data, and a model for flood risk assessment of power corridors, the slope factor is used to identify high-slope areas where the water flow velocity is relatively fast during floods and has a greater impact on power facilities; the distance factor is used to judge the distance between the utility pole and the water body. During floods, the closer the utility pole is to the water body, the more likely it is to be submerged and pose a risk of pole collapse; the flood risk factor is used to identify potential high-flood-risk areas based on the prediction results of the U-Net model.

[0138] Based on the above factors, risk assessment is performed on the input utility pole location data to obtain the specific locations of high-risk poles and the impact on power facilities. In the case of flood disasters, the present invention classifies the risk of utility pole collapse into 5 levels according to the distance between the utility pole and the water body and the slope. The risk level class judgment criteria are as follows:

[0139]

[0140] Among them, d is the distance between the utility pole point and the water body, and slope is the slope of this point.

[0141] Example 2, referring to Figures 4 - 7 , is the second embodiment of the present invention, which provides a method for flood risk assessment of power corridors based on deep learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0142] Taking Area A as the test site, Area A contains four divided regions, namely A1, A2, A3, and A4, with complex and diverse geographical conditions, including karst landforms, hilly mountains, and dense rivers and lakes. It has a subtropical monsoon climate with abundant precipitation and uneven seasonal distribution. Its rich vegetation and biodiversity increase the environmental complexity. Under such geographical and climatic conditions, there are many challenges in extracting water bodies by traditional methods. The limestone characteristics of karst landforms are easily confused with the water body spectrum, vegetation coverage and frequent floods make it difficult to identify water bodies, cloudy and rainy weather affects the clarity of remote sensing images, and complex terrain undulations increase the difficulty of identifying water body boundaries. These factors comprehensively lead to the complexity and challenges of accurately extracting and monitoring water bodies within Area A. To address these challenges, several partial regions within Area A were selected as the research objects, and Sentinel-1 satellite images were used as the basic data input. Combining slope information and power pole position data for auxiliary analysis, an effective assessment of the flood risk of power corridors was achieved through experimental applications of deep learning techniques.

[0143] First, according to the actual needs, download the Sentinel-1 VH polarization image data of Area A. Use the Digital Elevation Model (DEM) data to calculate the slope information and generate a slope distribution map. Obtain the power pole position data from the power department to ensure accurate data positioning.

[0144] In the second step, perform preprocessing on the Sentinel-1 images, including radiometric correction, geometric correction, noise removal, and image cropping. Extract all water bodies in these images based on visual interpretation and establish water body mask data. Finally, segment the original image and the water body mask data, and cut all the images into 256x256 image blocks. Rename the segmented images so that the original images and the water body masks correspond one by one, serving as the dataset for the subsequent deep learning method of extracting water bodies, as shown in the appendix Figure 4 shown.

[0145] In the third step, input the processed dataset into the established U-Net model, adopt the cross-entropy loss function and the Adam optimizer for model training. Monitor the loss value and accuracy during the training process, adjust the model parameters to prevent overfitting, as shown in the appendix Figure 5 shown. When the loss value and accuracy meet the requirements, save the trained model file for the next prediction.

[0146] In the fourth step, newly collect the images to be predicted, preprocess them, use the trained model to predict the new data in Area A, generate a water body distribution map of the input image area, and save it as a shp vector file, as shown in the appendix Figure 6 shown.

[0147] In the fifth step, calculate the accuracy of water body extraction and generate a table of the accuracy of water body extraction. Calculate the water body area in region A generated with the water body mask data generated by visual interpretation as the reference map. Calculate the MIOU, recall rate, and precision of the generated water body image. Finally, we obtained a detailed table of the accuracy of water body extraction, as shown in Table 1; the overlap between the generated water body and the visually interpreted water body is greater than 80%, and both the recall rate and precision are close to 1, which can meet the applications in reality.

[0148] Table 1 Table of the accuracy of water body extraction

[0149] Region MIOU Recall Precision Date Area A1 0.8790 0.8395 0.8925 2021.05.07 Area A2 0.9122 0.9287 0.8891 2021.05.12 Area A3 0.9221 0.9370 0.8985 2021.05.07 Area A4 0.8442 0.7317 0.9384 2021.01.06

[0150] In the sixth step, combine the slope data, power pole position data, and the model to conduct a flood risk assessment of the power corridor.

[0151] Slope factor, identify high-slope areas where the water flow velocity is relatively fast during floods, which has a greater impact on power facilities.

[0152] Distance factor, judge the distance between the power poles and the water body. During floods, the closer the power poles are to the water body, the more likely they are to be submerged, resulting in the risk of pole collapse.

[0153] Flood risk factor, based on the prediction results of the U-Net model, identify potential high-flood-risk areas.

[0154] According to the above factors, conduct a risk assessment on the input power pole position data to obtain the specific locations of high-risk power poles and the affected conditions of power facilities, as shown in the appendix Figure 7 and promptly propose protective measures and emergency plans for high-risk areas.

[0155] The above results verify the effectiveness of the flood risk assessment method for power corridors based on deep learning. This study, based on the U-Net deep learning model and combined with multi-source remote sensing images and geographic information data, has successfully achieved the accurate assessment of the flood risk of power corridors. The experimental results show that this method has high accuracy and robustness in water body extraction and risk identification. The research results can provide a scientific basis for disaster prevention and mitigation of power facilities and have important engineering application value. The flood risk assessment method for power corridors based on deep learning, compared with traditional methods, can exclude the influence of mountain shadows and terrain undulations and accurately extract water bodies from high-resolution Sentinel-1 data. Secondly, this method also adds a risk assessment module for power equipment, which can classify the importance of power facilities and pay more attention to the risk assessment of important power facilities. It can comprehensively consider factors such as slope, distance, and flood risk, calculate a comprehensive risk index, and more comprehensively assess the risk of power corridors.

[0156] Embodiment 3, the third embodiment of the present invention, which is different from the previous two embodiments in that:

[0157] If the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0158] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0159] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part with one or more wirings (electronic device), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0160] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0161] Example 4, referring to Figure 8 , which is the fourth embodiment of the present invention. This embodiment provides a power corridor flood risk assessment system based on deep learning, including a data processing module 10, a model construction module 20, an accuracy evaluation module 30, and a risk assessment module 40;

[0162] The data processing module 10 includes a data set construction unit 11 and an image annotation and adjustment unit 12;

[0163] Obtain radar images from the Sentinel-1 satellite, transfer the original radar images to a Speckle filter for noise removal, generate denoised images for geometric correction, locate the images, and transfer the corrected images to the radiation correction unit 11. Transfer the normalized images to the image annotation and adjustment unit 12 for manual or semi-automatic annotation of water bodies, generate annotation data, uniformly adjust the annotated images, output a training data set of a fixed size, and transfer the processed data set to the model construction module 20;

[0164] After the data processing module 10 completes data preprocessing, it transfers the processed training data to the model construction module 20.

[0165] The model construction module 20 includes a model selection unit 21, an encoder unit 22, a decoder unit 23, a skip connection unit 24, and an output and loss unit 25;

[0166] The model selection unit 21 selects a U-Net model for model construction. The encoder unit 22 receives the training data set from the data processing module 10, performs calculations on multiple convolutional layers and pooling layers to generate feature maps, and transfers the feature maps to the decoder unit 23 to perform upsampling and convolutional operations, gradually restoring the image resolution and generating intermediate output feature maps. The skip connection unit 24 splices the corresponding encoder feature maps and decoder feature maps at each decoding stage to enhance the details of the output. The output and loss unit 25 generates a segmentation result, calculates the loss function for the result, and feeds it back to the model selection unit 21;

[0167] The model construction module 20 trains the model based on the input data and sends the prediction results to the accuracy evaluation module 30.

[0168] The accuracy evaluation module 30 receives the segmentation results and the ground truth labels output by the model, generates a confusion matrix, calculates metrics based on the confusion matrix, and feeds the results back to the model construction module 20 to indicate whether the model needs to be retrained or optimized.

[0169] The accuracy evaluation module calculates the performance metrics of the model and feeds the results back to the model construction module 20 for subsequent optimization.

[0170] The risk assessment module 40 receives the water body extraction results, the distances between the utility pole positions and the water body, and the slope information, calculates the risk level, generates corresponding protection suggestions based on the risk assessment results, and records the corresponding relationship between the risk level and the suggestions.

[0171] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for assessing the flood risk of power corridors based on deep learning, characterized in that: including obtaining radar image data covering the power corridor and its surrounding areas by Sentinel-1 satellite and performing preprocessing; establishing a U-Net model to perform image segmentation on the preprocessed images and extract water body areas in the radar images; recollecting data and inputting it into the model, and evaluating the accuracy of water body extraction by calculating indicators such as confusion matrix, recall rate, precision, and MIOU.

2. The method for flood risk assessment of power corridors based on deep learning according to claim 1, wherein: The preprocessing includes using a Speckle filter to remove speckle noise in the radar images, selecting an appropriate filter window size to balance noise removal and image detail retention; converting the image pixel values to backscatter coefficients, and performing geometric correction on the images according to geographical factors and sensor characteristics to accurately position the images in space; performing radiometric correction to eliminate the radiometric differences caused by changes in sensor characteristics and observation conditions in the images, and making the image brightness and contrast consistent; cropping the images according to the geographical scope of the power corridor to extract the region of interest; performing normalization processing on the image data to standardize the pixel values to the range of [0, 255]; extracting all water bodies in the image data according to visual interpretation, establishing water body mask data, and segmenting the original image and the water body mask data.

3. The method for flood risk assessment of power corridors based on deep learning according to claim 2, characterized in that: The water body mask data includes, after preprocessing the Sentinel-1 (i.e., Sentinel-1) image, extracting all water bodies in the image according to visual interpretation. The interpreter manually identifies the water body areas by observing the gray scale, texture, and terrain features of the image, converts the water body areas into binary mask data, marks the water body areas in the mask as 1, and the non-water body areas as 0, establishes the water body mask data, segments the original image and the water body mask data, cuts the entire image into 256x256 image blocks, and renames the segmented images. The original image corresponds one-to-one with the water body mask, serving as the dataset for subsequent deep learning to extract water bodies.

4. The method for evaluating the flood risk of a power corridor based on deep learning according to claim 3, wherein: The U-Net model includes that the U-Net model performs image segmentation through a symmetric encoder and decoder structure; the encoder extracts local and global features of the image through a convolutional layer of 3x3 in each layer and a pooling layer of 2x2 max pooling, gradually increasing the abstraction degree of the features; the decoder part restores the image resolution through a deconvolution layer of 2x2 upsampling, combines skip connections to retain high-resolution features, and performs water body segmentation; converting the number of channels of the feature map to the required number of classes through a 1x1 convolutional layer, outputting the classification probability of each pixel, and distinguishing different classes of the model through a loss function to reduce misclassification.

5. The method for flood risk assessment of power corridors based on deep learning according to claim 4, wherein: The loss function includes adopting a cross-entropy loss function and an Adam optimizer to perform model training, iteratively adjusting the model parameters, and monitoring the loss value and accuracy rate during the training process; when the loss value is less than the set threshold 1 and the accuracy rate exceeds the set standard threshold 2, save the trained model file, and perform further prediction. Recollect the predicted images and perform preprocessing, use the trained model to predict the new data of the water body area, generate a water body distribution map of the input image area, and save it as a shp vector file.

6. The method for evaluating the flood risk of a power corridor based on deep learning according to claim 5, wherein: The evaluation of the accuracy of water body extraction includes calculating the accuracy of water body extraction, generating an extraction water body accuracy table, and calculating the water body area and the water body mask data generated by visual interpretation; Evaluate the accuracy of water body extraction by calculating the indicators of confusion matrix, recall rate, precision rate, and mean intersection over union; The recall rate Recal is: The precision rate Precision is: Among them, TP is true positive, FN is false negative, and FP is false positive; The mean intersection over union MIOU is: Among them, the calculation formula of IOU is as follows: Among them, N is the total number of categories; When the extraction accuracy meets the requirements, combine the extracted water body data with the location information and regional slope of the power bridge to conduct a flood risk assessment and evaluate the flood risk level of power equipment; According to the distance and slope of the power bridge from the water body, the risk level is divided into five levels, providing a scientific basis for the protection and management of power facilities.

7. The method for evaluating the flood risk of a power corridor based on deep learning according to claim 6, characterized in that: The risk level includes dividing the risk level into five levels according to the distance and slope of the power bridge from the water body: high risk, medium-high risk, medium risk, medium-low risk, and low risk; The risk level class judgment standard is: Among them, d is the distance of the power bridge from the water body, and slope is the slope of the power bridge from the water body.

8. A system adopting a method for flood risk assessment of a power corridor based on deep learning as described in any one of claims 1 to 7, characterized in that: It includes a data processing module, a model construction module, an accuracy evaluation module, and a risk assessment module; The data processing module includes a data set construction unit and an image annotation and adjustment unit; Obtain radar images from the Sentinel-1 satellite, transfer the original radar images to the Speckle filter for noise removal, generate denoised images for geometric correction, locate the images, and transfer the corrected images to the radiometric correction unit. Transfer the normalized images to the image annotation and adjustment unit for manual or semi-automatic annotation of water bodies, generate annotation data, uniformly adjust the annotated images, output a training data set of a fixed size, and transfer the processed data set to the model construction module; The model construction module includes a model selection unit, an encoder unit, a decoder unit, a skip connection unit, and an output and loss unit; The model selection unit selects the U-Net model for model construction. The encoder unit receives the training data set from the data processing module, performs calculations on multiple convolutional layers and pooling layers to generate feature maps, and transfers the feature maps to the decoder unit to perform upsampling and convolutional operations, gradually restoring the image resolution and generating intermediate output feature maps. The skip connection unit splices the corresponding encoder feature maps and decoder feature maps at each decoding stage to enhance the details of the output. The output and loss unit generates a segmentation result, calculates the loss function for the result, and feeds it back to the model selection unit; The accuracy evaluation module receives the segmentation result and the true label output by the model, generates a confusion matrix, calculates the indicators based on the confusion matrix, and feeds the result back to the model construction module to indicate whether the model needs to be retrained or optimized; The risk assessment module receives the water body extraction results, the distance between the pole positions and the water body, and the information of the slope, calculates the risk level, generates corresponding protection suggestions according to the results of the risk assessment, and records the corresponding relationship between the risk level and the suggestions.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for flood risk assessment of a power corridor based on deep learning described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for flood risk assessment of a power corridor based on deep learning described in any one of claims 1 to 7.

Images