Power transmission line forest fire risk assessment method and system based on image processing, and medium

Abstract

The invention discloses a power transmission line forest fire risk assessment method and system based on image processing and a medium, and belongs to the field of image processing, and the method comprises the steps: obtaining a field image when a power transmission line has a forest fire, and carrying out the image preprocessing of the field image, and obtaining a preprocessed image; inputting the preprocessed image into a pre-trained feature recognition model, so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps of different levels, and predicting the first feature maps through a region proposal network in the feature recognition model to determine a corresponding proposal region, outputting a discharge arc image based on the first feature map and the proposed region; according to the method, the curve contour corresponding to the discharge arc image is extracted, the corresponding arc parameter is determined based on the curve contour, and the power transmission line forest fire risk is evaluated based on the arc parameter, so that the accuracy of power transmission line forest fire risk evaluation can be improved.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method, system and medium for evaluating wildfire risk of power transmission lines based on image processing. Background Art

[0002] When a wildfire occurs, the complex combustion environment can trigger arcing around transmission lines. This arcing not only poses a serious threat to the safe and stable operation of the transmission lines, but can also cause line faults and tripping, impacting the reliability of power supply. Therefore, identifying the characteristics of wildfire arcing can more accurately assess the risk to transmission lines, allowing for proactive preventive measures to reduce fire risks.

[0003] Identifying arc characteristics during wildfire discharges primarily relies on traditional electrical detection methods and simple image observation techniques. While electrical detection methods can monitor changes in electrical parameters during the discharge process, they struggle to intuitively capture detailed features such as the arc's spatial shape and position. This is particularly true in complex wildfire environments, where monitoring of the arc's initiation and progression is incomplete. Image observation methods, however, often struggle to accurately extract arc characteristics when faced with interfering factors such as strong light, dense smoke, and complex terrain at a wildfire site, leading to potential misjudgments or omissions. Therefore, identifying arc characteristics during wildfire discharges is crucial to improving the accuracy of wildfire risk assessments for transmission lines. Summary of the Invention

[0004] The present application provides a method, system and medium for transmission line wildfire risk assessment based on image processing to accurately identify the characteristics of wildfire discharge arcs, thereby improving the accuracy of transmission line wildfire risk assessment.

[0005] The present invention provides a method for assessing the risk of wildfires in power transmission lines based on image processing, comprising:

[0006] Acquire an on-site image when a wildfire occurs on a power transmission line, and perform image preprocessing on the on-site image to obtain a preprocessed image;

[0007] Inputting the preprocessed image into a pretrained feature recognition model so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predicting the first feature maps using a region proposal network in the feature recognition model to determine corresponding proposed regions, and then outputting a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network;

[0008] A curve profile corresponding to the discharge arc image is extracted, corresponding arc parameters are determined based on the curve profile, and the wildfire risk of the transmission line is evaluated based on the arc parameters.

[0009] The embodiment of the present application performs image preprocessing on the on-site image to ensure that the image input to the feature recognition model is clear and has high contrast, thereby helping the model to more accurately identify the discharge arc features; in this way, the backbone network in the feature recognition model performs convolution processing on the preprocessed image, which can capture the underlying features such as edges and textures in the image as well as more advanced semantic features, helping the model to more accurately identify the discharge arc features, thereby more accurately assessing the potential threat of wildfires to transmission lines; through the region proposal network for prediction, a series of proposed regions that may contain discharge arcs can be quickly generated, which narrows the scope of subsequent detailed analysis, improves computational efficiency, helps subsequent steps focus on the real discharge arc region, avoids misjudgment of background or other irrelevant areas, and improves the accuracy of risk assessment; by extracting the curve contour corresponding to the discharge arc image, the arc parameters can be accurately calculated, thereby achieving an accurate assessment of the transmission line wildfire risk. Compared with the prior art, the present application can accurately identify the characteristics of wildfire discharge arcs, thereby improving the accuracy of transmission line wildfire risk assessment.

[0010] Furthermore, the predicting of the first feature map by the region proposal network in the feature recognition model to determine the corresponding proposed region is specifically as follows:

[0011] The region proposal network performs convolution processing on the first feature map to obtain local features, and adjusts the number of channels of the local features to obtain a second feature map and a third feature map;

[0012] Calculate the binary classification probability of each position in the second feature map by using the first activation function, wherein the binary classification probability is the probability that each position belongs to the target or the background;

[0013] Predicting the offset of the bounding box at each position in the third feature map by a first bounding box regression algorithm;

[0014] A corresponding proposed region is determined based on the binary classification probability and the offset.

[0015] In this way, prediction through the region proposal network can quickly generate a series of proposed regions that may contain discharge arcs, narrowing the scope of subsequent detailed analysis, improving computational efficiency, and helping subsequent steps focus on the actual discharge arc area, avoiding misjudgment of background or other irrelevant areas, thereby improving the accuracy of risk assessment.

[0016] Furthermore, the feature recognition model includes a fully connected layer and a fully convolutional network, and then outputs a discharge arc image based on the first feature map and the proposed area, specifically:

[0017] Determine a target feature map based on the first feature map and the proposed region;

[0018] Inputting the target feature map into the fully connected layer and the fully convolutional network respectively, predicting the target position through the fully connected layer in combination with a second bounding box regression algorithm, and predicting the target category through the fully connected layer in combination with a second activation function;

[0019] Perform pixel-by-pixel prediction through the fully convolutional network prediction to determine the target mask;

[0020] A discharge arc image is determined based on the target position, the target category, and the target mask.

[0021] In this way, by combining the first feature map with the proposed area to obtain the target feature map, the discharge arc can be located more accurately and its key features can be extracted to accurately determine the discharge arc image, providing a basis for subsequent arc parameter calculation.

[0022] Furthermore, the extracting of the curve profile corresponding to the discharge arc image is specifically as follows:

[0023] performing grayscale processing on the discharge arc image to obtain a third image, and removing noise in the third image by filtering to obtain a fourth image;

[0024] performing image enhancement on the fourth image to obtain a fifth image;

[0025] Edge information in the fifth image is detected by an edge detection algorithm, and a corresponding curve profile is determined based on the edge information.

[0026] In this way, by extracting the curve profile corresponding to the discharge arc image, the arc parameters can be accurately calculated, thereby achieving an accurate assessment of the wildfire risk of the transmission line.

[0027] Furthermore, the arc parameters include arc length, and the corresponding arc parameters are determined based on the curve profile, specifically:

[0028] Determine the conversion factor from pixels to actual length based on a preset rule;

[0029] The pixel distance between pixel points in the curve profile is determined based on the conversion factor, and the pixel distance between all adjacent edge points on the curve profile is accumulated by digital integration to determine the arc length.

[0030] In this way, by analyzing the curve profile, the arc length can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks on transmission lines, making the assessment results more accurate and reliable.

[0031] Furthermore, the arc parameters also include arc width, and the corresponding arc parameters are determined based on the curve profile, specifically:

[0032] establishing a rectangular coordinate system for the discharge arc image;

[0033] Traversing each horizontal straight line in the rectangular coordinate system, and calculating the number of pixels occupied by the intersection of each horizontal straight line and the discharge arc image;

[0034] Based on the number of pixels and the conversion factor, a corresponding arc width is obtained.

[0035] In this way, by analyzing the curve profile, the arc width can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks of transmission lines, making the assessment results more accurate and reliable.

[0036] Furthermore, the arc parameters include arc brightness, and the corresponding arc parameters are determined based on the curve profile, specifically:

[0037] Extracting all pixel points in the curve profile and counting the grayscale values ​​corresponding to the pixel points to obtain the pixel grayscale value;

[0038] The average brightness and the brightness variance corresponding to the curve profile are determined based on the grayscale values ​​of the pixel points, wherein the arc brightness includes the average brightness and the brightness variance.

[0039] In this way, by analyzing the curve profile, the arc brightness can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks on transmission lines, making the assessment results more accurate and reliable.

[0040] Furthermore, the on-site image is preprocessed to obtain a preprocessed image, specifically:

[0041] Performing grayscale transformation on the scene image to obtain a first image;

[0042] removing noise from the first image by using a median filtering algorithm to obtain a second image;

[0043] The second image is input into a preset image defogging model to remove smoke from the second image to obtain a preprocessed image.

[0044] In this way, by performing image preprocessing on the on-site images, it can be ensured that the images input to the feature recognition model are clear and have high contrast, which helps the model to more accurately identify the discharge arc characteristics.

[0045] Another embodiment of the present invention further provides a transmission line wildfire risk assessment system based on image processing, comprising: an acquisition module, an identification module, and an assessment module;

[0046] The acquisition module is used to acquire an on-site image when a wildfire occurs on a power transmission line, and perform image preprocessing on the on-site image to obtain a preprocessed image;

[0047] The recognition module is configured to input the preprocessed image into a pretrained feature recognition model, so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predict the first feature maps through a region proposal network in the feature recognition model to determine corresponding proposed regions, and then output a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network;

[0048] The evaluation module is used to extract the curve profile corresponding to the discharge arc image, determine the corresponding arc parameters based on the curve profile, and evaluate the wildfire risk of the transmission line based on the arc parameters.

[0049] The embodiment of the present application performs image preprocessing on the on-site image to ensure that the image input to the feature recognition model is clear and has high contrast, thereby helping the model to more accurately identify the discharge arc features; in this way, the backbone network in the feature recognition model performs convolution processing on the preprocessed image, which can capture the underlying features such as edges and textures in the image as well as more advanced semantic features, helping the model to more accurately identify the discharge arc features, thereby more accurately assessing the potential threat of wildfires to transmission lines; through the region proposal network for prediction, a series of proposed regions that may contain discharge arcs can be quickly generated, which narrows the scope of subsequent detailed analysis, improves computational efficiency, helps subsequent steps focus on the real discharge arc region, avoids misjudgment of background or other irrelevant areas, and improves the accuracy of risk assessment; by extracting the curve contour corresponding to the discharge arc image, the arc parameters can be accurately calculated, thereby achieving an accurate assessment of the transmission line wildfire risk. Compared with the prior art, the present application can accurately identify the characteristics of wildfire discharge arcs, thereby improving the accuracy of transmission line wildfire risk assessment.

[0050] Another embodiment of the present invention further provides a computer-readable storage medium item, comprising: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to execute the steps of the transmission line wildfire risk assessment method based on image processing as described in the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 This is a flow chart of an embodiment of a method for assessing wildfire risk of power transmission lines based on image processing provided by the present application;

[0053] Figure 2 is a schematic diagram of the on-site image provided by this application;

[0054] Figure 3 This is a schematic diagram of the breakdown test platform layout provided by this application;

[0055] Figure 4 Schematic diagram of the model structure of the feature recognition model provided by this application;

[0056] Figure 5 is a schematic diagram of a discharge arc image provided by this application;

[0057] Figure 6 It is the arc morphology diagram under the coordinate axis provided by this application;

[0058] Figure 7 This is a structural diagram of an embodiment of a transmission line wildfire risk assessment system based on image processing provided in this application. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0061] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0062] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0063] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0064] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0065] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0066] Arc discharges on transmission lines caused by wildfires can threaten line safety, cause faults and tripping, and disrupt power supply. Identifying the characteristics of wildfire arc discharges is crucial for risk assessment and fire prevention. Traditional electrical detection methods can monitor changes in electrical parameters, but they struggle to comprehensively monitor the initiation and progression of arc discharges. Image observation methods are also susceptible to interference from the wildfire scene, leading to misjudgments or omissions. Therefore, the key to improving the accuracy of wildfire risk assessments on transmission lines lies in accurately identifying the characteristics of wildfire arc discharges.

[0067] Next, we will analyze the main nouns involved in this application:

[0068] Wildfire arc characteristics refer to the specific attributes and manifestations of discharge arcs in images that can be identified and analyzed when wildfires cause discharges on transmission lines. These characteristics can be used to identify the existence of arcs, assess their risks, and analyze their impact on transmission lines.

[0069] Mask-RCNN (Mask Region-based Convolutional Neural Network) is an advanced object detection and instance segmentation model, mainly used to handle complex scene understanding tasks in computer vision. It aims to simultaneously perform object detection and pixel-level instance segmentation, that is, while identifying each target object in the image, it can accurately outline the contours of each object.

[0070] Next, the technical solution of the present invention is introduced:

[0071] See also Figure 1 In order to solve the problem that the existing technology cannot accurately extract arc features and thus cannot accurately assess the wildfire risk of transmission lines, an embodiment of the present invention provides a method for assessing the wildfire risk of transmission lines based on image processing, including: steps S101 to S103;

[0072] Step S101, obtaining a scene image when a wildfire occurs on a power transmission line, and performing image preprocessing on the scene image to obtain a preprocessed image;

[0073] In some embodiments, on-site images of a wildfire occurring on a power transmission line are obtained. Specifically, multiple high-definition cameras are installed in areas where a wildfire is prone to occur on the power transmission line. When a wildfire occurs, images of the wildfire scene on the power transmission line are collected in real time through the cameras, i.e., the on-site images of the present invention. The collected on-site images are transmitted to an image processing system, wherein a schematic diagram of the on-site images is shown in Figure 2.

[0074] It should be noted that in addition to high-definition cameras, infrared thermal imaging equipment can penetrate thick smoke and more effectively capture the heat information generated by the discharge arc, which may provide more reliable image data in a thick smoke environment. Therefore, it is possible to consider combining equipment with infrared thermal imaging capabilities with visible light cameras to collect images. However, the problem of fusion of infrared and visible light images needs to be solved subsequently to ensure the effective use of comprehensive information.

[0075] In this way, through multi-view high-definition camera acquisition and optimized image preprocessing technology, the interference of environmental factors at the wildfire scene on the image is effectively reduced, the image quality is improved, the arc features can be presented more clearly, and the feasibility and accuracy of feature extraction are enhanced.

[0076] In some embodiments, the image preprocessing of the scene image to obtain a preprocessed image includes: performing grayscale transformation on the scene image to obtain a first image; removing noise in the first image through a median filtering algorithm to obtain a second image; inputting the second image into a preset image defogging model to remove smoke in the second image to obtain a preprocessed image. Specifically, first, the color scene image is converted into a grayscale image to obtain the first image, and the contrast of the image is enhanced to make the difference between the arc and the background more obvious, in preparation for subsequent processing; secondly, a suitable median filter window size is selected, such as 3×3, 5×5, etc., and each pixel in the first image is median filtered with the determined window size. In each window, the grayscale values ​​of the pixels are arranged in ascending order, and the grayscale value of the middle position is taken as the new grayscale value of the center pixel of the window, thereby removing noise points such as salt and pepper noise in the image to obtain the second image; finally, a deep learning-based image dehazing model is trained in advance, and the second image is input into the preset image dehazing model. The model will automatically analyze and process the smoke features in the image, and correct and remove the smoke part in the image through the learned feature mapping relationship, and finally output the preprocessed image after smoke removal, thereby restoring the clarity of the image.

[0077] It should be noted that the grayscale transformation method can be the average method, weighted average method, etc. The specific grayscale transformation algorithm is not the focus of this application and will not be expanded.

[0078] It should be noted that the median filtering algorithm can effectively suppress noise interference caused by factors such as combustion particles without destroying the edge information of the image. The specific median filtering algorithm is not the focus of this application and will not be expanded.

[0079] It should be noted that the selection of the median filter window size needs to be determined by comprehensively considering the intensity of the noise and the detailed characteristics of the image. If the window is too large, the image details may be blurred, while if it is too small, the noise may not be effectively removed.

[0080] It should be noted that the image dehazing model can be a dehazing model based on a convolutional neural network (CNN). The model is trained by a large number of images with smoke and their corresponding smoke-free images to learn the characteristic representation and removal method of smoke. By inputting the second image into the trained image dehazing model, the smoke in the second image can be directly removed to obtain a preprocessed image.

[0081] In this way, by performing image preprocessing on the on-site images, it can be ensured that the images input to the feature recognition model are clear and have high contrast, which helps the model to more accurately identify the discharge arc characteristics.

[0082] Step S102: inputting the preprocessed image into a pretrained feature recognition model so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predicting the first feature maps through a region proposal network in the feature recognition model to determine corresponding proposed regions, and then outputting a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network;

[0083] In some embodiments, the feature recognition model is a model that has been pre-trained using a large amount of wildfire discharge arc image data, so that the feature recognition model can automatically learn arc feature patterns and effectively distinguish arcs from other similar interference objects.

[0084] It should be noted that the feature recognition model can be a network model based on Mask-RCNN. In addition, you can try to use recurrent neural networks (RNN) and its variants (such as long short-term memory networks LSTM, gated recurrent units GRU) to process the time series data of arc features. Especially when analyzing the dynamic change process of arcs, RNN may be able to better capture the temporal dependency of features, but the training complexity of the model may be relatively high.

[0085] It should be noted that the training process for the Mask-RCNN-based feature recognition model is as follows: First, several sets of breakdown tests under vegetation fire gaps are set up to collect training images. A DC voltage generator generates high voltage, which is controlled by adjusting a protective resistor and a voltage divider. A woodpile is ignited to generate flames, simulating a wildfire environment. Arc images are captured using DV1 and DV2. A large amount of image data containing different types of arcs and various interference scenarios is also collected. Second, annotation tools are used to annotate the arcs in the training image set, generating bounding boxes and class labels. The images are then grayscale transformed, median filtered, and dehazed to improve image quality and clarify arc features. Then, Mask-RCNN is used as the feature recognition model. The preprocessed image data is fed into the Mask-RCNN model, which performs forward and backward propagation, automatically updating the model parameters. Visualization tools such as TensorBoard are used to monitor the training loss and accuracy, and the model's progress is monitored. Checkpoints are saved regularly for subsequent training or testing. Finally, the model is evaluated using the validation set, and indicators such as the loss value and accuracy on the validation set are calculated. When the loss value is minimized, the training is completed and a trained feature recognition model is obtained.

[0086] It should be noted that the breakdown test platform layout diagram is as follows Figure 3As shown, the DC voltage generator on the left is connected to a protective resistor and a voltage divider to generate and regulate DC high voltage and control discharge conditions. The DC voltage generator is connected to a simulated wire, which is fixed to a rope via an insulator and suspended from a crane. At the other end, the simulated wire is connected to a metal plate placed on a flame generation platform. A scale is installed next to the metal plate to measure parameters such as arc length. The flame generation platform is located below the metal plate, and a woodpile is placed on the flame generation platform to create a wildfire environment. DV1 and DV2 are installed on either side of the metal plate to capture and record images and data of the discharge arc.

[0087] It should be noted that when installing the camera, it should be ensured that the camera's installation position and angle can fully cover the area where discharge arcs may occur. At the same time, the camera should have high resolution, high frame rate, and wide dynamic range to adapt to the complex lighting conditions at the wildfire scene. At the same time, the camera's installation position and angle should be optimized according to the on-site environment and the direction of the simulated wires to minimize the terrain's obstruction to shooting and ensure that clear arc images can be obtained.

[0088] In some embodiments, the model structure diagram of the feature recognition model is as follows: Figure 4 As shown in the figure, the feature recognition model includes a backbone network (ResNet 50 / 101+FPN), a region proposal network (RPN), a region of interest alignment (RolAlign), a fully convolutional network (Fully Convolution Net) and three branches of fully connected layers (FC layers).

[0089] It should be noted that the feature recognition model can not only learn the visual features of arcs in images, but also combine these quantified feature parameters to more accurately determine whether the object in the image is an arc, thereby achieving more accurate arc recognition and segmentation.

[0090] In some embodiments, the preprocessed image is input into a pretrained feature recognition model so that the backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps of different levels. Specifically, the preprocessed image is input into the feature recognition model. At this time, the preprocessed image will enter the backbone network composed of ResNet 50 or ResNet 101 and FPN. The backbone network will perform a series of convolution operations on the preprocessed image to extract feature information at different levels and output the first feature map (Feature Map). This process is achieved by sliding the convolution kernel on the image to perform convolution operations, and continuously extracting the bottom-level to high-level features such as the edges and textures of the preprocessed image.

[0091] In some embodiments, the prediction of the first feature map by the region proposal network in the feature recognition model to determine the corresponding proposed region includes: the region proposal network performs convolution processing on the first feature map to obtain local features, and adjusts the number of channels of the local features respectively to obtain a second feature map and a third feature map; calculates the binary classification probability of each position in the second feature map by a first activation function, wherein the binary classification probability is the probability that each position belongs to the target or the background; predicts the offset of the bounding box at each position in the third feature map by a first bounding box regression algorithm; and determines the corresponding proposed region based on the binary classification probability and the offset. Specifically, first, after receiving the first feature map, the region proposal network uses a convolution kernel (for example, a size of 3×3) to slide on the first feature map, and performs a weighted sum operation on the pixels at each position and its neighborhood to capture the local correlation in the first feature map and generate a new feature map, i.e., a local feature. Next, two convolution kernels (e.g., 1×1) are used to convolve the local features. These convolution kernels (e.g., 1×1) do not change the spatial size, but can adjust the number of channels. For example, if the number of channels of a local feature is 256, 1×1 convolution can adjust it to different numbers, such as 128 (to obtain the second feature map) and 256 (to obtain the third feature map), to meet the needs of subsequent branch processing. A Softmax function (also known as the first activation function) is then applied to the second feature map to calculate the probability of each location in the second feature map belonging to the target or background (the sum of the two is 1), thereby obtaining a binary classification probability. Since the third feature map represents the default anchor box relative to that location, the first bounding box regression algorithm is used to predict the offset of the bounding box in the x and y directions. Finally, according to a set probability threshold (e.g., 0.5), candidate regions belonging to the target category are screened from the binary classification probability results of the second feature map. The bounding box offsets predicted by the third feature map are then used to adjust the bounding boxes of these candidate regions to obtain more accurate proposed region locations and sizes.

[0092] It should be noted that in order to remove redundant and overlapping proposals, non-maximum suppression is also required on the adjusted proposals. The specific steps are: sort the candidate regions from high to low according to their probability values, retain the regions with the highest probability, and suppress other regions whose overlap exceeds a certain threshold (for example, 0.7), ultimately obtaining a set of high-quality, non-overlapping proposals.

[0093] It should be noted that the first, second, and third do not indicate the order of precedence, but can be understood as nouns. The first feature map is a feature map generated by performing a series of convolution operations on the preprocessed image through the backbone network, and the second feature map and the third feature map are feature maps obtained by adjusting the number of channels corresponding to the local features extracted by the first convolution layer.

[0094] In this way, prediction through the region proposal network can quickly generate a series of proposed regions that may contain discharge arcs, narrowing the scope of subsequent detailed analysis, improving computational efficiency, and helping subsequent steps focus on the actual discharge arc area, avoiding misjudgment of background or other irrelevant areas, thereby improving the accuracy of risk assessment.

[0095] In some embodiments, the feature recognition model includes a fully connected layer and a fully convolutional network, and then outputs a discharge arc image based on the first feature map and the proposed region, including: determining a target feature map based on the first feature map and the proposed region; inputting the target feature map into the fully connected layer and the fully convolutional network respectively, predicting the target position through the fully connected layer combined with a second bounding box regression algorithm, and predicting the target category through the fully connected layer combined with a second activation function; performing pixel-by-pixel prediction through the fully convolutional network prediction to determine the target mask; determining the discharge arc image based on the target position, the target category and the target mask. Specifically, first, after determining the first feature map and the proposed region, the first feature map and the proposed region need to be input into the region of interest alignment (RolAlign) together. The region of interest alignment (RolAlign) extracts fixed-size features on the first feature map according to the proposed region, and maps the features corresponding to the proposed regions of different sizes to a uniform size through methods such as bilinear interpolation to obtain a target feature map for subsequent processing. Secondly, the target feature map is flattened into a one-dimensional vector and input into the fully connected layer and the fully convolutional network respectively. Then, the feature vector output by the fully connected layer is processed by the second bounding box regression algorithm to predict the coordinate offset and size adjustment parameters of the target bounding box to determine the target position (determine the position of the target in the image). At the same time, the feature vector output by the fully connected layer is classified and predicted by the second activation function (such as softmax), and the probability distribution of each target category is calculated to obtain the prediction result of the target category (determine what object the target belongs to). After that, the fully convolutional network will predict the probability of each pixel belonging to the target pixel by pixel, and generate a mask prediction map of the same size as the input feature map, that is, determine the target mask to facilitate arc segmentation, wherein these mask images clearly show the arc part separated from the flame image. Finally, according to the predicted target position, target category and target mask, the mask corresponding to the target position in the mask prediction map is extracted, and combined with the target category information to generate the final discharge arc image, wherein the schematic diagram of the discharge arc image is as shown Figure 5 shown.

[0096] It should be noted that for each pixel, the fully convolutional network will output the confidence that the pixel belongs to the target. By setting an appropriate threshold, pixels with confidence higher than the threshold are judged as pixels in the target area, and finally a complete target mask is formed to achieve pixel-level segmentation of the target.

[0097] In this way, the feature recognition model can effectively distinguish arcs from other similar interference objects (such as burning branches, fire reflections, etc.) by automatically learning the arc feature patterns from a large number of samples.

[0098] It should be noted that the fully connected layer performs linear transformation and nonlinear activation on the input feature vector to extract high-level semantic features. The fully convolutional network processes the feature map through convolution operations to extract features while maintaining spatial information.

[0099] In this way, by combining the first feature map with the proposed area to obtain the target feature map, the discharge arc can be located more accurately and its key features can be extracted to accurately determine the discharge arc image, providing a basis for subsequent arc parameter calculation.

[0100] Step S103 : extracting a curve profile corresponding to the discharge arc image, determining corresponding arc parameters based on the curve profile, and evaluating the wildfire risk of the transmission line based on the arc parameters.

[0101] In some embodiments, extracting the curve profile corresponding to the discharge arc image includes: grayscale processing the discharge arc image to obtain a third image, removing noise from the third image by filtering to obtain a fourth image; performing image enhancement on the fourth image to obtain a fifth image; detecting edge information in the fifth image using an edge detection algorithm, and determining the corresponding curve profile based on the edge information. Specifically, first, because the main feature of the arc is reflected in brightness, the grayscale value of each pixel is calculated using a common grayscale conversion formula such as a weighted average method to convert the discharge arc image into a grayscale image, reduce the data dimension, and highlight the brightness information of the image to obtain the third image, wherein the relevant calculation formula is: Gray = 0.299 × R + 0.587 × G + 0.114 × B, where Gray is the grayscale value, and R, G, and B are the three-channel values ​​of the color discharge arc image. Secondly, the third image is filtered using a selected filtering method (such as Gaussian filtering, median filtering, etc.) to obtain the fourth image. Then, based on the quality and characteristics of the fourth image, a suitable image enhancement method is selected, such as histogram equalization, contrast enhancement, etc., to obtain the fifth image. This approach can expand the dynamic range of the image grayscale, enhance the image contrast, make the difference between the arc and the background more obvious, and facilitate subsequent contour extraction. Afterwards, the fifth image is edge detected using a selected edge detection algorithm (such as the Canny operator, Sobel operator, Prewitt operator, etc.) to obtain edge information in the image. For example, the Canny operator first performs Gaussian smoothing on the image, then calculates the gradient amplitude and direction, and then performs non-maximum suppression. Finally, the true edge point is determined by a double threshold algorithm to obtain preliminary edge information of the arc. Finally, based on the detected edge information, a contour tracking algorithm (such as a tracking algorithm based on pixel neighborhood relationships) is used to connect the edge points to form a closed contour, that is, to determine the corresponding curve contour.

[0102] It should be noted that Gaussian filtering is suitable for removing Gaussian noise, while median filtering is suitable for removing salt and pepper noise. Gaussian filtering performs a weighted average of image pixels based on a Gaussian function, smoothing the image, reducing noise interference, and making the arc area clearer. For a pixel with coordinates I(x,y) in the image, its filtered grayscale value I′(x,y) is the weighted sum of the grayscale values ​​of the pixels in the original image centered at that point, using the Gaussian function.

[0103] In this way, by extracting the curve profile corresponding to the discharge arc image, the arc parameters can be accurately calculated, thereby achieving an accurate assessment of the wildfire risk of the transmission line.

[0104] In some embodiments, the arc parameters include arc length, and determining the corresponding arc parameters based on the curve profile includes: determining a conversion factor from pixels to actual length based on a preset rule; determining the pixel distance between pixels in the curve profile based on the conversion factor, and accumulating the pixel distances between all adjacent edge points on the curve profile by digital integration to determine the arc length. Specifically, first, a reference object with a known actual length is selected in the image, such as a simulated wire-board gap length of x meters. After image processing and calculation, it is known that the number of pixels occupied by the gap length is N, and the actual length occupied by one pixel is (x / N) meters (performance conversion factor). Secondly, on the curve profile, the actual distance of each pixel can be calculated using the conversion factor. Thereafter, an edge detection algorithm (such as the Canny operator) and a contour extraction algorithm (such as a tracking algorithm based on pixel neighborhood relationships) are used to obtain a pixel coordinate sequence of the curve profile, and then the pixel distances between all adjacent edge points on the curve profile are accumulated by digital integration to calculate the arc length, wherein the calculation formula for the arc length is: Where, L is the arc length; (x i ,y i ) and (x i+1 ,y i+1 ) are two consecutive points on the curve, and N is the total number of points on the curve.

[0105] In this way, by analyzing the curve profile, the arc length can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks on transmission lines, making the assessment results more accurate and reliable.

[0106] In some embodiments, the arc parameters also include arc width. The arc parameters corresponding to the arc are determined based on the curve profile, specifically as follows: establishing a rectangular coordinate system for the discharge arc image; traversing each horizontal straight line in the rectangular coordinate system, and calculating the number of pixels occupied by the intersection of each horizontal straight line and the discharge arc image; and obtaining the corresponding arc width based on the number of pixels and the conversion factor. Specifically, first, a rectangular coordinate system is established with the arc top of the curve profile as the origin and the connected straight line as the x-axis, wherein the arc morphology diagram under the coordinate axis is as follows: Figure 6 As shown, then, in the range of y = 0 to y = N, traverse each horizontal straight line (y is a constant) and calculate the number of pixels occupied by the intersection of each straight line and the discharge arc image; finally, multiply the number of pixels where each straight line intersects the discharge arc image by the actual length occupied by each pixel (that is, the conversion factor) to obtain the arc width at the y coordinate.

[0107] In this way, by analyzing the curve profile, the arc width can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks of transmission lines, making the assessment results more accurate and reliable.

[0108] In some embodiments, the arc parameters include arc brightness, and determining the corresponding arc parameters based on the curve profile is specifically as follows: specifically, extracting all pixels in the curve profile and counting the grayscale values ​​corresponding to each pixel to obtain the pixel grayscale value; and determining the average brightness and brightness variance corresponding to the curve profile based on the pixel grayscale values, wherein the arc brightness includes the average brightness and the brightness variance. First, extract all pixels in the arc region from the curve profile and collect the grayscale values ​​of all pixels in the arc region to obtain the pixel grayscale value; then, calculate the average grayscale value of all pixels in the arc region to obtain the average brightness, and calculate the variance of the grayscale values ​​of all pixels in the arc region to obtain the brightness variance.

[0109] In some embodiments, the average brightness reflects the overall luminous intensity of the arc area. A higher average brightness indicates a stronger arc. Assume that there are M pixels in the arc area, and their grayscale values ​​are g1, g2, ..., g M , then the calculation formula for the average brightness is Where, is the average brightness, g i is the brightness of the i-th pixel. The brightness variance reflects the discreteness of the grayscale value in the arc area. A smaller brightness variance indicates that the arc is more uniform and has better stability; a larger brightness variance indicates that the arc is uneven, and there may be flickering or other unstable factors. The calculation formula of the brightness variance is Where, where σ2 is the brightness variance, is the average brightness, g i is the brightness of the i-th pixel.

[0110] In this way, by analyzing the curve profile, the arc brightness can be accurately calculated, providing a quantitative basis for the assessment of wildfire risks on transmission lines, making the assessment results more accurate and reliable.

[0111] This feature extraction and recognition method that combines traditional image processing and deep learning can comprehensively and accurately extract key features such as the geometry and brightness of the arc, and effectively distinguish the arc from interference through the deep learning model, significantly improving the accuracy and reliability of feature recognition.

[0112] It should be noted that in order to ensure accuracy, the model output results need to be verified using the features obtained by contour extraction. If the arc bounding box output by the model is significantly different from the actual range of the arc obtained by contour extraction, the bounding box output by the model can be fine-tuned based on the contour features, or the contour features can be used as additional input features and fed back into the model for secondary training or optimization, thereby continuously improving the accuracy of arc detection and identification by the feature recognition model.

[0113] This method combines traditional image processing methods with deep learning techniques. For arc geometric features such as length, width, and shape, an edge detection algorithm (such as Canny edge detection) is used to extract the arc's outline. Morphological processing (such as dilation and erosion) is then used to optimize the outline information. The arc's geometric parameters are then calculated based on the outline coordinates. For brightness feature extraction, the grayscale distribution of the arc region in the image is analyzed, and parameters such as average brightness and brightness variance are calculated to characterize the arc's luminous intensity and stability.

[0114] In some embodiments, the wildfire risk of transmission lines is assessed based on the arc parameters. Specifically, after arc feature extraction, the extracted geometric and brightness parameters are analyzed and quantified. For example, arc status levels corresponding to different parameter ranges are set. For example, when the arc length is less than a certain threshold and the average brightness is low, the arc state is determined to be low-risk; when the arc length exceeds a certain threshold and the brightness variance is large, the arc state is determined to be high-risk.

[0115] It should be noted that when the transmission line fire risk level reaches a certain level, different control methods are automatically adopted based on the fire risk level. For example, in a low-risk arc state, ventilation equipment can be controlled to operate at a lower power to improve air circulation and prevent heat accumulation caused by the arc. The monitoring equipment's shooting frequency can be adjusted, and the shooting interval can be appropriately reduced to reduce data storage pressure. In a high-risk arc state, the alarm device will be immediately activated, sounding an audible and visual alarm to alert relevant personnel; the power supply related to the arcing area will be automatically cut off to avoid possible electrical failures or fires; and the power of the ventilation equipment will be increased to accelerate heat dissipation and remove any harmful gases that may be generated.

[0116] It should be noted that the extracted arc characteristic parameters (arc length, arc width and arc brightness) and classification results (whether it is an arc) can also be integrated to draw arc characteristic curves (such as arc length change curve over time, width change curve, etc.), and the position and shape changes of the arc can be displayed in the identification results to provide comprehensive and accurate decision-making basis for transmission line operation and maintenance personnel. Such analysis and visualization of the identification results realizes the intuitive presentation and dynamic monitoring of the discharge arc characteristics, which helps operation and maintenance personnel to timely understand the impact of wildfires on transmission lines, provide strong support for taking corresponding protection and repair measures, and improve the operation safety of transmission lines during wildfires.

[0117] The embodiment of the present application performs image preprocessing on the on-site image to ensure that the image input to the feature recognition model is clear and has high contrast, thereby helping the model to more accurately identify the discharge arc features; in this way, the backbone network in the feature recognition model performs convolution processing on the preprocessed image, which can capture the underlying features such as edges and textures in the image as well as more advanced semantic features, helping the model to more accurately identify the discharge arc features, thereby more accurately assessing the potential threat of wildfires to transmission lines; through the region proposal network for prediction, a series of proposed regions that may contain discharge arcs can be quickly generated, which narrows the scope of subsequent detailed analysis, improves computational efficiency, helps subsequent steps focus on the real discharge arc region, avoids misjudgment of background or other irrelevant areas, and improves the accuracy of risk assessment; by extracting the curve contour corresponding to the discharge arc image, the arc parameters can be accurately calculated, thereby achieving an accurate assessment of the transmission line wildfire risk. Compared with the prior art, the present application can accurately identify the characteristics of wildfire discharge arcs, thereby improving the accuracy of transmission line wildfire risk assessment.

[0118] like Figure 7 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0119] An embodiment of the present invention provides a transmission line wildfire risk assessment system based on image processing, comprising: an acquisition module, an identification module, and an assessment module;

[0120] The acquisition module is used to acquire an on-site image when a wildfire occurs on a power transmission line, and perform image preprocessing on the on-site image to obtain a preprocessed image;

[0121] The recognition module is configured to input the preprocessed image into a pretrained feature recognition model, so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predict the first feature maps through a region proposal network in the feature recognition model to determine corresponding proposed regions, and then output a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network;

[0122] The evaluation module is used to extract the curve profile corresponding to the discharge arc image, determine the corresponding arc parameters based on the curve profile, and evaluate the wildfire risk of the transmission line based on the arc parameters.

[0123] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, which can implement the image processing-based transmission line wildfire risk assessment method provided by any of the above-mentioned method embodiments of the present invention.

[0124] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort.

[0125] Based on the above-mentioned embodiment of the transmission line wildfire risk assessment method based on image processing, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the transmission line wildfire risk assessment method based on image processing according to any embodiment of the present invention is implemented.

[0126] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0127] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0128] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0129] Based on the above method embodiments, another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the image processing-based transmission line wildfire risk assessment method described in any one of the above method embodiments of the present invention.

[0130] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0131] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for assessing the risk of wildfires in power transmission lines based on image processing, characterized in that: include: Acquire an on-site image when a wildfire occurs on a power transmission line, and perform image preprocessing on the on-site image to obtain a preprocessed image; Inputting the preprocessed image into a pretrained feature recognition model so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predicting the first feature maps using a region proposal network in the feature recognition model to determine corresponding proposed regions, and then outputting a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network; A curve profile corresponding to the discharge arc image is extracted, corresponding arc parameters are determined based on the curve profile, and the wildfire risk of the transmission line is evaluated based on the arc parameters.

2. The method for assessing the risk of wildfires on power transmission lines based on image processing according to claim 1, characterized in that: The predicting of the first feature map by the region proposal network in the feature recognition model to determine the corresponding proposed region is specifically as follows: The region proposal network performs convolution processing on the first feature map to obtain local features, and adjusts the number of channels of the local features to obtain a second feature map and a third feature map; Calculate the binary classification probability of each position in the second feature map by using the first activation function, wherein the binary classification probability is the probability that each position belongs to the target or the background; Predicting the offset of the bounding box at each position in the third feature map by a first bounding box regression algorithm; A corresponding proposed region is determined based on the binary classification probability and the offset.

3. The method for assessing the risk of wildfires on power transmission lines based on image processing according to claim 1, characterized in that: The feature recognition model includes a fully connected layer and a fully convolutional network, and then outputs a discharge arc image based on the first feature map and the proposed area, specifically: Determine a target feature map based on the first feature map and the proposed region; Inputting the target feature map into the fully connected layer and the fully convolutional network respectively, predicting the target position through the fully connected layer in combination with a second bounding box regression algorithm, and predicting the target category through the fully connected layer in combination with a second activation function; Perform pixel-by-pixel prediction through the fully convolutional network prediction to determine the target mask; A discharge arc image is determined based on the target position, the target category, and the target mask.

4. The method for assessing transmission line wildfire risk based on image processing according to claim 1, characterized in that: The extracting of the curve profile corresponding to the discharge arc image is specifically as follows: performing grayscale processing on the discharge arc image to obtain a third image, and removing noise in the third image by filtering to obtain a fourth image; performing image enhancement on the fourth image to obtain a fifth image; Edge information in the fifth image is detected by an edge detection algorithm, and a corresponding curve profile is determined based on the edge information.

5. The method for assessing transmission line wildfire risk based on image processing according to claim 1, characterized in that: The arc parameters include arc length, and the corresponding arc parameters are determined based on the curve profile, specifically: Determine the conversion factor from pixels to actual length based on a preset rule; The pixel distance between pixel points in the curve profile is determined based on the conversion factor, and the pixel distance between all adjacent edge points on the curve profile is accumulated by digital integration to determine the arc length.

6. The method for assessing the risk of wildfires on power transmission lines based on image processing according to claim 5, characterized in that: The arc parameters also include arc width, and the corresponding arc parameters are determined based on the curve profile, specifically: establishing a rectangular coordinate system for the discharge arc image; Traversing each horizontal straight line in the rectangular coordinate system, and calculating the number of pixels occupied by the intersection of each horizontal straight line and the discharge arc image; Based on the number of pixels and the conversion factor, a corresponding arc width is obtained.

7. The method for assessing the risk of wildfires on power transmission lines based on image processing according to claim 5, characterized in that: The arc parameters include arc brightness, and the corresponding arc parameters are determined based on the curve profile, specifically: Extracting all pixel points in the curve profile and counting the grayscale values ​​corresponding to the pixel points to obtain the pixel grayscale value; The average brightness and the brightness variance corresponding to the curve profile are determined based on the grayscale values ​​of the pixel points, wherein the arc brightness includes the average brightness and the brightness variance.

8. The method for assessing transmission line wildfire risk based on image processing according to claim 1, characterized in that: The image preprocessing is performed on the scene image to obtain a preprocessed image, specifically: Performing grayscale transformation on the scene image to obtain a first image; removing noise from the first image by using a median filtering algorithm to obtain a second image; The second image is input into a preset image defogging model to remove smoke from the second image to obtain a preprocessed image.

9. A transmission line wildfire risk assessment system based on image processing, characterized in that: include: Acquisition module, identification module and assessment module; The acquisition module is used to acquire an on-site image when a wildfire occurs on a power transmission line, and perform image preprocessing on the on-site image to obtain a preprocessed image; The recognition module is configured to input the preprocessed image into a pretrained feature recognition model, so that a backbone network in the feature recognition model performs convolution processing on the preprocessed image to obtain first feature maps at different levels, and predict the first feature maps through a region proposal network in the feature recognition model to determine corresponding proposed regions, and then output a discharge arc image based on the first feature maps and the proposed regions, wherein the feature recognition model includes the backbone network and the region proposal network; The evaluation module is used to extract the curve profile corresponding to the discharge arc image, determine the corresponding arc parameters based on the curve profile, and evaluate the wildfire risk of the transmission line based on the arc parameters.

10. A computer-readable storage medium, characterized in that include: A stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to perform the steps of the transmission line wildfire risk assessment method based on image processing according to any one of claims 1 to 8.

Images