Transformer substation fire intelligent monitoring and early warning method and system
By obtaining infrared and visible light images in the substation and using YOLO algorithm and autoencoder for flame monitoring, the problem of fire monitoring and early warning of substations is solved, and rapid identification and early warning is achieved, and damage and costs are reduced.
Patent Information
- Application Number
- CN202510032447.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-09
- Publication Date
- 2025-04-29
AI Technical Summary
Due to the remote distribution and changing weather environment, substations are prone to rapid development of fires due to high temperatures, and it is difficult for existing technology to effectively monitor and early warning fires.
An intelligent monitoring and early warning method for substation fire is adopted. By acquiring infrared data and visible light images, integrating data and using the YOLO algorithm to extract image features, distinguishing flame and smoke images, and combining autoencoder and wavelet transformation for noise reduction processing.
It realizes rapid identification and early warning of substation fires, reduces the damage to facilities by fires, reduces the cost of maintenance and equipment replacement, and improves economic benefits.
Smart Images
Figure CN119904812A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of substation monitoring technology, and specifically to a method and system for intelligent monitoring and early warning of substation fires. Background Art
[0002] With the continuous development and progress of society, the number of substations built in my country has increased significantly, and the exposed safety issues cannot be ignored. The safety production needs of substations and the national standards for substation monitoring require the development of intelligent fire monitoring and early warning systems for substations. Most substations are located in remote areas with changeable weather conditions. Substations are easily affected by high temperatures, which leads to rapid development of fires. With the rapid development of artificial intelligence and computer vision technology, fire monitoring systems combining multimodal imaging technology and deep learning are gradually becoming a research hotspot. How to monitor and warn substations for fires based on image recognition is a technical problem that needs to be solved in this field. Summary of the invention
[0003] In view of this, the present application provides a method and system for intelligent monitoring and early warning of substation fires, which can perform fire monitoring and early warning on substations.
[0004] In a first aspect, the present application provides an intelligent monitoring and early warning method for substation fires, comprising: obtaining infrared data and visible light images of objects during inspection of the substation; extracting temperature data based on the infrared data and obtaining an infrared image; fusing the infrared image and the visible light image to obtain fused data; extracting image features from the fused data based on the YOLO algorithm, and distinguishing them into flame images and smoke images based on the image features, and predicting the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box; using an autoencoder neural network to learn and reconstruct the flame image and the smoke image for reflection denoising; and denoising the denoised flame image and the smoke image using wavelet transform and SVD.
[0005] In combination with the first aspect, in a possible implementation, after extracting temperature data based on the infrared data and obtaining an infrared image, the method also includes: converting the infrared image from the spatial domain to the frequency domain to obtain frequency domain data through Fourier wavelet decomposition; quantizing the subband coefficients of the frequency domain data, compressing the data, and enhancing important features; entropy encoding the quantized subband coefficients; and performing wavelet reconstruction on the entropy-coded frequency domain data, and converting it to the spatial domain to obtain the processed infrared image.
[0006] In combination with the first aspect, in a possible implementation manner, the method further includes: detecting a failed pixel in the infrared image; and estimating and replacing the failed pixel using a reference value of a neighboring pixel of the failed pixel.
[0007] In combination with the first aspect, in a possible implementation, the extracting image features from the fused data based on the YOLO algorithm, distinguishing them into flame images and smoke images according to the image features, and predicting the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box include: extracting the image features from the fused data through the Backbone layer; using the Neck layer to deeply extract and process the image features; and using multiple parallel convolutional layers of the Head layer to distinguish the image features into the flame image and the smoke image, and obtain the bounding boxes, coordinates and sizes of the flame image and the smoke image.
[0008] In combination with the first aspect, in a possible implementation method, the use of multiple parallel convolution layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtain the bounding boxes, coordinates, and sizes of the flame image and the smoke image includes: dividing the parallel convolution layer into a classification convolution layer and a bounding box regression convolution layer; setting the number of output channels of the classification convolution layer to 2, classifying the image features, and determining the flame image and the smoke image; and setting the number of output channels of the bounding box regression convolution layer to 4 times the maximum number of anchor points to obtain output data of the flame image and the smoke image, wherein the output data includes a bounding box, coordinates, size, and confidence.
[0009] In combination with the first aspect, in a possible implementation, the using of multiple parallel convolutional layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtaining the bounding boxes, coordinates, and sizes of the flame image and the smoke image also includes: using non-maximum suppression to eliminate overlapping bounding boxes; filtering the output data according to a preset confidence threshold; marking a first bounding box and a first label on the flame image; and marking a second bounding box and a second label on the smoke image.
[0010] In combination with the first aspect, in a possible implementation method, the use of multiple parallel convolutional layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtaining the bounding boxes, coordinates and sizes of the flame image and the smoke image also includes: loading a real flame image and a real smoke image; comparing the predicted flame image, the smoke image with the real flame image and the real smoke image to evaluate the prediction accuracy.
[0011] In combination with the first aspect, in a possible implementation method, extracting the image features from the fused data through the Backbone layer includes: preloading preset features; calculating the correlation between the preset features and other features in the fused data; adjusting feature weights according to the correlation; if the correlation is greater than a preset correlation value, increasing the feature weight of the preloaded features; and extracting the image features corresponding to the preset features from the fused data based on the feature weights.
[0012] In combination with the first aspect, in a possible implementation, the use of an autoencoder neural network to learn and reconstruct the flame image and the smoke image for reflection noise reduction includes: compressing the original data of the flame image and the smoke image into low-dimensional codes through an encoder; expanding and reconstructing the low-dimensional codes through a decoder to obtain reconstructed data; comparing the reconstruction errors of the original data and the reconstructed data; calculating the gradient parameters of the reconstruction error using an error back propagation algorithm; and updating the network weights in proportion to the gradient parameters to train to reduce the reconstruction error.
[0013] In a second aspect, the present application provides an intelligent monitoring and early warning system for substation fires, comprising: a data acquisition module, configured to: acquire infrared data and visible light images of objects during inspection of the substation; extract temperature data based on the infrared data and obtain an infrared image; fuse the infrared image and the visible light image to obtain fused data; a YOLO detection module, communicatively connected to the data acquisition module, the YOLO detection module configured to: extract image features from the fused data based on a YOLO algorithm, and distinguish them into flame images and smoke images based on the image features, and predict the coordinates and size of the flame image bounding box, and the coordinates and size of the smoke image bounding box; and a denoising module, communicatively connected to the data acquisition module and the YOLO detection module respectively, the denoising module configured to: use an autoencoder neural network to learn and reconstruct the flame image and the smoke image for reflection denoising; and use wavelet transform and SVD to denoise the denoised flame image and the smoke image. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 The figure is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by an embodiment of the present application.
[0015] Figure 2 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0016] Figure 3 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0017] Figure 4 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0018] Figure 5 FIG. 4 is a schematic diagram of a hierarchy of a YOLO algorithm provided in another embodiment of the present application.
[0019] Figure 6 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0020] Figure 7 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0021] Figure 8 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0022] Fig. 9 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0023] Fig.10 Shown is a schematic diagram of the method steps of a substation fire intelligent monitoring and early warning method provided by another embodiment of the present application.
[0024] Fig.11 The figure shows a schematic diagram of the system structure of a substation fire intelligent monitoring and early warning system provided in one embodiment of the present application. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0026] Figure 1 FIG. 1 is a schematic diagram of a method for intelligent monitoring and early warning of substation fire according to an embodiment of the present application. Figure 1 As shown, the present application provides a substation fire intelligent monitoring and early warning method, the method comprising: Step 110: Acquire infrared data and visible light images of objects during inspection of the substation.
[0027] In this step, when inspecting the substation, the infrared data of each object in the substation is obtained using an infrared thermal imaging camera, and the object is photographed using the inspection camera to obtain a visible light image of the object.
[0028] Step 120: extract temperature data based on the infrared data and obtain an infrared image.
[0029] In this step, the infrared data is converted into temperature data and sent to the background system to be processed into an infrared image.
[0030] Step 130: Fuse the infrared image and the visible light image to obtain fused data.
[0031] In this step, the infrared image data and the visible light image data are combined into a data group to obtain fused data for subsequent use.
[0032] Step 140 : extracting image features from the fused data based on the YOLO algorithm, distinguishing the image into a flame image and a smoke image according to the image features, and predicting the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box.
[0033] In this step, the YOLO algorithm extracts flame features and smoke features from visible light images based on preset parameters in a single forward propagation, thereby distinguishing flame images from smoke images, and simultaneously predicting the categories and position parameters of multiple target objects, achieving an optimal balance between speed and accuracy. Step 150: Use an autoencoder neural network to learn and reconstruct the flame image and the smoke image to perform reflection denoising.
[0034] In this step, the autoencoder, as an unsupervised learning neural network, is good at learning the implicit representation of input data and reconstructing the input data. Through training, the autoencoder can learn how to reconstruct the noise-free original image from the noisy image and effectively eliminate noise interference such as reflections.
[0035] Step 160: De-noise the flame image and smoke image using wavelet transform and SVD.
[0036] In this step, traditional machine learning techniques of wavelet transform and singular value decomposition (SVD) are used for noise reduction and denoising. Wavelet transform can capture the characteristics of signals at different times and frequencies through multi-scale analysis, perform scaling and translation operations on the signals, and achieve multi-scale refinement. Specifically, wavelet transform decomposes the flame image and smoke image into components of different scales, and performs threshold processing on these components to remove or reduce noise components, and finally reconstruct a clear signal. SVD is a denoising method that decomposes a matrix into the product of three specific matrices. Specifically, SVD can effectively eliminate noise in flame images and smoke images by retaining the most important singular values, while ensuring that the main features of the flame image and smoke image are retained.
[0037] This embodiment can quickly discover and predict the coordinates and size of the flame image boundary box and the coordinates and size of the smoke image boundary box by identifying the flame image and the smoke image and performing anti-noise denoising on the inspection image of the substation, that is, predict the direction of the flame and smoke, which is helpful to warn of the fire, reduce the damage of the fire to the substation facilities, reduce the maintenance and equipment replacement costs, and improve the economic benefits.
[0038] Figure 2 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Figure 2 As shown, after step 120, the substation fire intelligent monitoring and early warning method further includes: Step 121: Convert the infrared image from the spatial domain to the frequency domain by Fourier wavelet decomposition to obtain frequency domain data.
[0039] Step 122: quantize the sub-band coefficients of the frequency domain data, compress the data, and enhance important features.
[0040] Step 123: Perform entropy coding on the quantized sub-band coefficients.
[0041] Step 124: Perform wavelet reconstruction based on the frequency domain data after entropy coding, and convert it into the spatial domain to obtain a processed infrared image.
[0042] In this embodiment, the infrared image is first converted from the spatial domain to the frequency domain by Fourier wavelet decomposition, and the spatial and frequency information of the image is processed. Then, the sub-band coefficients obtained by decomposition are quantized to compress the image data and enhance important features. Then, entropy coding is performed on the quantized sub-band coefficients to reduce the amount of data and optimize storage and transmission efficiency. Finally, wavelet reconstruction is performed based on the entropy-coded data to restore the image, ensuring that key information is retained.
[0043] Figure 3FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Figure 3 As shown, the substation fire intelligent monitoring and early warning method also includes: Step 125: Detect failed pixels in the infrared image.
[0044] Step 126: Use the reference values of the neighboring pixels of the failed pixel to estimate and replace the failed pixel.
[0045] In this embodiment, the reference value is the average value of neighboring pixels or a reference value obtained by other algorithms. To address the problem of failed pixels that may be encountered during infrared imaging, this embodiment performs Bad Pixel Replacement (BPR) processing to eliminate the adverse effects of these pixels on image quality, provide clearer and more accurate images, and achieve more efficient monitoring and analysis.
[0046] Figure 4 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Figure 4 As shown, step 140 includes: Step 141: extract image features from the fused data through the Backbone layer.
[0047] Step 142: Use the Neck layer to deeply extract and process image features.
[0048] Step 143: using multiple parallel convolutional layers of the Head layer, distinguish the image features into flame images and smoke images, and obtain the bounding boxes, coordinates, and sizes of the flame images and smoke images.
[0049] Figure 5 FIG. 2 is a schematic diagram of the hierarchy of the YOLO algorithm provided by another embodiment of the present application. Figure 5 The YOLO algorithm includes Backbone layer, Neck layer, and Head layer. The fused data is imported into the Backbone layer for processing, including extracting image features. The introduction of the improved backbone network and neck architecture in YOLO can significantly improve the efficiency of feature extraction, enhance the accuracy of object detection and performance in complex scenes. The CF2 module in YOLO is upgraded to C3K2, which improves the efficiency of feature extraction. After the SPPF module, YOLO introduced the C2PSA module to enhance the model's ability to refine features. The Neck module is responsible for in-depth extraction and processing of features. The introduction of the C3K2 module in YOLO realizes the effective integration of multi-scale features and enhances the recognition ability of targets of various sizes.
[0050] The head layer consists of multiple parallel convolutional layers, which are responsible for predicting the coordinates, size, and target category of the image bounding box. YOLO uses deep separable convolution technology to reduce the amount of calculation and improve the operating efficiency of the model. YOLO processes the convolutional layers of detection and classification tasks separately. The number of channels of the feature map output by the convolutional layer of bounding box regression is 4 times the maximum number of anchor points, and the number of channels of the classification convolutional layer is determined according to the number of target categories. In order to reduce the amount of calculation, deep separable convolution is used in classification tasks to replace traditional convolution operations.
[0051] In terms of feature training and feature optimization for flame and smoke images, YOLO redesigned the loss function to train the network more effectively and balance multiple key objectives in the target detection task. The loss function calculation is divided into two main parts: classification and regression. The objectness branch is no longer included. The classification part continues to use binary cross entropy loss, while the regression part uses distribution focus loss combined with CIoU loss.
[0052] In summary, the YOLO algorithm can handle flame detection and smoke detection in complex scenes, and can take into account detection accuracy, processing speed and model robustness. Visible light images provide rich texture information, while infrared images provide temperature information. Fusion of visible light images and infrared images can enhance the accuracy of the model's detection and positioning of flames and smoke.
[0053] Figure 6 FIG. 2 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Specifically, Figure 6 As shown, step 143 includes: Step 1431, split the parallel convolution layer into a classification convolution layer and a bounding box regression convolution layer.
[0054] Step 1432: Set the number of output channels of the classification convolution layer to 2, classify the image features, and determine the flame image and the smoke image.
[0055] Step 1433: Set the number of output channels of the bounding box regression convolution layer to 4 times the maximum number of anchor points, and obtain output data of the flame image and the smoke image, the output data including the bounding box, coordinates, size, and confidence.
[0056] Figure 7 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Figure 7 As shown, step 143 also includes: Step 1434: Use non-maximum suppression to eliminate overlapping bounding boxes.
[0057] Step 1435: Filter the output data according to a preset confidence threshold.
[0058] Step 1436: Mark a first bounding box and a first label on the flame image.
[0059] Step 1437: Mark a second bounding box and a second label on the smoke image.
[0060] In this embodiment, the output data is post-processed, including applying non-maximum suppression to eliminate overlapping bounding boxes and filtering the final detection results according to the confidence threshold. The processed results will draw bounding boxes and category labels on the original image to intuitively display the detected objects.
[0061] Figure 8 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Figure 8 As shown, step 143 also includes: Step 1438: Load the real flame image and the real smoke image.
[0062] Step 1439: compare the predicted flame image and smoke image with the real flame image and real smoke image to evaluate the prediction accuracy.
[0063] In this embodiment, the real flame image and the real smoke image are obtained by real-time shooting. Since real images are provided, that is, real labels are provided, the prediction accuracy performance of the model can be evaluated, including calculation accuracy, recall rate and other indicators. Finally, the detection results can be output in the form of images, videos or text reports to complete the entire target detection process.
[0064] Fig. 9 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Fig. 9 As shown, step 141 includes: Step 1411, preload preset features.
[0065] Step 1412: Calculate the correlation between the preset feature and other features in the fused data.
[0066] Step 1413: Adjust feature weights according to the degree of association.
[0067] Step 1414: If the correlation degree is greater than a preset correlation value, the feature weight of the preloaded feature is increased.
[0068] Step 1415: extract image features corresponding to preset features from the fused data based on feature weights.
[0069] In this embodiment, the preset features are key features. By calculating the weight distribution of the preset features, it is possible to strengthen important information and ignore irrelevant content, thereby improving the model's ability to recognize the preset features. This process involves evaluating the correlation between each preset feature and all other features, and dynamically adjusting the feature weights based on these evaluation results, so that the model can focus on the key features that are most relevant to the current task, ultimately achieving more accurate prediction and classification.
[0070] Fig.10 FIG. 1 is a schematic diagram of the steps of a method for intelligent monitoring and early warning of a substation fire provided by another embodiment of the present application. Fig.10 As shown, step 150 includes: Step 151: compress the original data of the flame image and the smoke image into low-dimensional codes through an encoder.
[0071] Step 152: Expand and reconstruct the low-dimensional code through a decoder to obtain reconstructed data.
[0072] Step 153: Compare the reconstruction errors of the original data and the reconstructed data.
[0073] Step 154: Calculate the gradient parameter of the reconstruction error using the error back propagation algorithm.
[0074] Step 155: Update the network weights in proportion to the gradient parameters to train and reduce the reconstruction error.
[0075] In this embodiment, the autoencoder is a neural network for unsupervised learning, which is good at learning the implicit representation of input data and reconstructing the input data. It is mainly composed of an encoder and a decoder. The encoder compresses the data into low-dimensional codes, while the decoder reconstructs these codes back to the original data. Through training, the autoencoder can learn how to reconstruct the original noise-free image from the noisy image, effectively eliminating noise interference such as reflections.
[0076] Fig.11 The figure shows a schematic diagram of the system structure of a substation fire intelligent monitoring and early warning system provided by an embodiment of the present application. The present application also provides a substation fire intelligent monitoring and early warning system, such as Fig.11 As shown, the system includes: a data acquisition module 101, a YOLO detection module 102 and a noise reduction module 103.
[0077] The data acquisition module 101 is configured to: acquire infrared data and visible light images of objects during inspection of the substation; extract temperature data based on the infrared data and obtain an infrared image; and fuse the infrared image and the visible light image to obtain fused data.
[0078] The YOLO detection module 102 is in communication connection with the data acquisition module 101, and the YOLO detection module 102 is configured to extract image features from the fused data based on the YOLO algorithm, and distinguish the image into a flame image and a smoke image according to the image features, and predict the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box.
[0079] The denoising module 103 is respectively connected to the data acquisition module 101 and the YOLO detection module 102 for communication. The denoising module 103 is configured to: use an autoencoder neural network to learn and reconstruct the flame image and the smoke image to perform reflection denoising; and use wavelet transform and SVD to denoise the denoised flame image and smoke image.
[0080] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.
[0081] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.
[0082] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.
[0083] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features of the present invention.
[0084] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A method for intelligent monitoring and early warning of substation fire, characterized in that: include: Acquire infrared data and visible light images of objects during substation inspections; Extract temperature data according to the infrared data and obtain an infrared image; fusing the infrared image and the visible light image to obtain fused data; Extracting image features from the fused data based on the YOLO algorithm, distinguishing the image into a flame image and a smoke image according to the image features, and predicting the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box; Using an autoencoder neural network to learn and reconstruct the flame image and the smoke image to perform reflection denoising; as well as The flame image and the smoke image after denoising are denoised by using wavelet transform and SVD.
2. The intelligent monitoring and early warning method for substation fire according to claim 1 is characterized in that: After extracting temperature data according to the infrared data and obtaining an infrared image, the method further includes: Converting the infrared image from the spatial domain to the frequency domain by Fourier wavelet decomposition to obtain frequency domain data; quantizing the subband coefficients of the frequency domain data, compressing the data, and enhancing important features; performing entropy coding on the quantized subband coefficients; and Wavelet reconstruction is performed on the frequency domain data after entropy coding, and the data is converted into the spatial domain to obtain the processed infrared image.
3. The intelligent monitoring and early warning method for substation fire according to claim 2 is characterized in that: Also includes: detecting failed pixels in the infrared image; as well as The failed pixel is estimated and replaced by using reference values of neighboring pixels of the failed pixel.
4. The intelligent monitoring and early warning method for substation fire according to claim 1 is characterized in that: The method of extracting image features from the fused data based on the YOLO algorithm, distinguishing the image into a flame image and a smoke image according to the image features, and predicting the coordinates and size of the flame image bounding box and the coordinates and size of the smoke image bounding box includes: Extracting the image features from the fused data through a Backbone layer; Using the Neck layer to deeply extract and process the image features; and The image features are distinguished as the flame image and the smoke image by using multiple parallel convolutional layers of the Head layer, and the bounding boxes, coordinates, and sizes of the flame image and the smoke image are obtained.
5. The intelligent monitoring and early warning method for substation fire according to claim 4 is characterized in that: The using of multiple parallel convolutional layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtaining the bounding boxes, coordinates, and sizes of the flame image and the smoke image includes: Splitting the parallel convolutional layer into a classification convolutional layer and a bounding box regression convolutional layer; Setting the number of output channels of the classification convolution layer to 2, classifying the image features, and determining the flame image and the smoke image; and The number of output channels of the bounding box regression convolution layer is set to 4 times the maximum number of anchor points, and output data of the flame image and the smoke image are obtained, wherein the output data includes a bounding box, coordinates, size, and confidence.
6. The intelligent monitoring and early warning method for substation fire according to claim 5 is characterized in that: The using multiple parallel convolution layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtaining the bounding boxes, coordinates, and sizes of the flame image and the smoke image also includes: Eliminate overlapping bounding boxes using non-maximum suppression; Filtering the output data according to a preset confidence threshold; marking a first bounding box and a first label on the flame image; and A second bounding box and a second label are marked on the smoke image.
7. The intelligent monitoring and early warning method for substation fire according to claim 6 is characterized in that: The using multiple parallel convolution layers of the Head layer to distinguish the image features as the flame image and the smoke image, and obtaining the bounding boxes, coordinates, and sizes of the flame image and the smoke image also includes: Load real flame images and real smoke images; The predicted flame image and smoke image are compared with the real flame image and real smoke image to evaluate the prediction accuracy.
8. The intelligent monitoring and early warning method for substation fire according to claim 4 is characterized in that: The extracting the image features from the fused data through the Backbone layer includes: Preloaded with preset features; Calculating the correlation between the preset feature and other features in the fused data; Adjust feature weights based on relevance; If the correlation degree is greater than a preset correlation value, increasing the feature weight of the preloaded feature; and Based on the feature weight, the image feature corresponding to the preset feature is extracted from the fused data.
9. The intelligent monitoring and early warning method for substation fire according to claim 5 is characterized in that: The using of an autoencoder neural network to learn and reconstruct the flame image and the smoke image to perform reflection denoising includes: Compressing the original data of the flame image and the smoke image into low-dimensional codes through an encoder; The low-dimensional code is expanded and reconstructed by a decoder to obtain reconstructed data; comparing reconstruction errors of the original data and the reconstructed data; Calculating the gradient parameter of the reconstruction error using an error back propagation algorithm; and The network weights are updated proportionally to the gradient parameters to train and reduce the reconstruction error.
10. An intelligent monitoring and early warning system for substation fire, characterized in that: include: The data acquisition module is configured to: acquire infrared data and visible light images of objects during inspection of the substation; Extract temperature data according to the infrared data and obtain an infrared image; fuse the infrared image and the visible light image to obtain fused data; A YOLO detection module is communicatively connected to the data acquisition module, and the YOLO detection module is configured to: extract image features from the fused data based on a YOLO algorithm, distinguish the image into a flame image and a smoke image according to the image features, and predict the coordinates and size of the flame image boundary box and the coordinates and size of the smoke image boundary box; and A denoising module is respectively connected to the data acquisition module and the YOLO detection module, and the denoising module is configured to: use an autoencoder neural network to learn and reconstruct the flame image and the smoke image to perform reflection denoising; and use wavelet transform and SVD to denoise the denoised flame image and the smoke image.