Substation online fire monitoring method, device, system and storage medium

Abstract

This application provides a method, device, system, and storage medium for online fire monitoring of substations. The method includes: preprocessing a first monitoring video of the substation to obtain a second monitoring video; when the brightness of an image frame in the second monitoring video is greater than or equal to a brightness threshold, reducing the background brightness of that image frame to obtain a third monitoring video; extracting static image features representing the presence or absence of smoke and flames from the second monitoring video, and extracting dynamic image features representing the amplitude of light change from the third monitoring video; and processing the static and dynamic image features using a fire identification model to obtain a fire identification result. This solution uses a fire identification model to analyze real-time photographs of substation equipment to detect whether a fire has occurred in the substation. Compared with the physical quantities monitored by fire detectors, the photographs taken by the camera equipment are almost unaffected by the outdoor environment, therefore this solution has higher accuracy.

Description

Technical Field

[0001] This invention relates to the field of substation operation and maintenance technology, and in particular to a method, device, system and storage medium for online fire monitoring of substations. Background Technology

[0002] Electricity is a primary energy source in people's lives today and a crucial prerequisite for ensuring stable socio-economic development and people's production and daily life. Substations, as hubs of the power grid, are key to ensuring the normal operation of society. Therefore, fire monitoring of substations is of paramount importance.

[0003] The current mainstream method for fire monitoring in substations is to install multiple fire detectors within the substation area. These detectors monitor physical quantities such as temperature, smoke, gas, and radiation intensity within the substation, and use these physical quantities to determine whether a fire has occurred.

[0004] The problem with this method is that the equipment in the substation is all set up in the open air, and the physical quantities monitored by the fire detectors are easily affected by environmental factors, so the accuracy of this method is low.

[0005] Based on the above, this invention patent is proposed: a substation fire monitoring online system. This system combines an improved convolutional neural network algorithm with multi-feature recognition, further improving the accuracy of fire monitoring and exhibiting high stability. It enables timely early warning and control in the early stages of a fire, greatly reducing fire-related losses, improving efficiency, and effectively enhancing the work efficiency and management level of maintenance personnel. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention provides a method, device, system, and storage medium for online fire monitoring of substations, thereby offering a more accurate substation fire monitoring solution.

[0007] The first aspect of this application provides a method for online fire monitoring of a substation, including:

[0008] Obtain the first monitoring video of the substation;

[0009] The first monitoring video is preprocessed to obtain the second monitoring video; wherein, the video preprocessing includes image cropping and image filtering;

[0010] Determine whether the brightness of the image frame in the second monitoring video is greater than or equal to a preset brightness threshold;

[0011] When the brightness of the image frame is greater than or equal to the brightness threshold, the background brightness of each image frame in the second monitoring video is reduced to obtain the third monitoring video;

[0012] Extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and extract dynamic image features representing the amplitude of light change from the third monitoring video;

[0013] The static image features and the dynamic image features are processed using a pre-trained fire identification model to obtain fire identification results; wherein, the fire identification model includes a static sub-model for processing the static image features and a dynamic sub-model for processing the dynamic image features;

[0014] When the fire identification result indicates that a fire has occurred, the control and early warning equipment outputs fire early warning information.

[0015] Optionally, the step of reducing the background brightness of each image frame in the second monitoring video to obtain the third monitoring video includes:

[0016] Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video.

[0017] The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

[0018] Optionally, after the control and early warning device outputs fire early warning information, it further includes:

[0019] Activate at least one fire-fighting device installed in the substation.

[0020] Optionally, the process of training the fire identification model includes:

[0021] Multiple training samples are obtained; each training sample includes a sample monitoring video, the static image features and dynamic image features corresponding to the sample monitoring video, and the pre-labeled real fire identification result, which indicates whether a fire has occurred or not.

[0022] The static and dynamic image features of each training sample are processed using the neural network model to be trained to obtain the predicted fire identification result corresponding to each training sample; wherein, the predicted fire identification result is either a fire has occurred or no fire has occurred.

[0023] The model loss of the neural network model to be trained is determined based on the deviation between the actual fire identification result and the predicted fire identification result for each sample.

[0024] Determine whether the model loss meets the preset convergence condition;

[0025] If the model loss does not meet the convergence condition, update the parameters of the neural network model to be trained according to the model loss, and return to the step of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample, until the model loss meets the convergence condition.

[0026] If the model loss satisfies the convergence condition, the neural network model to be trained is determined as a fire identification model.

[0027] The second aspect of this application provides an online fire monitoring device for substations, comprising:

[0028] The acquisition unit is used to acquire the first monitoring video of the substation.

[0029] A preprocessing unit is used to perform video preprocessing on the first monitoring video to obtain a second monitoring video; wherein, the video preprocessing includes image cropping and image filtering;

[0030] The judgment unit is used to determine whether the brightness of the image frame in the second monitoring video is greater than or equal to a preset brightness threshold.

[0031] An adjustment unit is used to reduce the background brightness of each image frame in the second monitoring video when the brightness of the image frame is greater than or equal to the brightness threshold, so as to obtain a third monitoring video.

[0032] The extraction unit is used to extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and to extract dynamic image features representing the amplitude of light change from the third monitoring video.

[0033] The identification unit is used to process the static image features and the dynamic image features using a pre-trained fire identification model to obtain a fire identification result; wherein, the fire identification model includes a static sub-model for processing the static image features and a dynamic sub-model for processing the dynamic image features;

[0034] The execution unit is used to control the early warning device to output fire early warning information when the fire identification result indicates that a fire has occurred.

[0035] Optionally, when the adjustment unit reduces the background brightness of each image frame in the second monitoring video to obtain the third monitoring video, it is specifically used for:

[0036] Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video.

[0037] The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

[0038] Optionally, after the execution unit controls the early warning device to output fire early warning information, it is further used for:

[0039] Activate at least one fire-fighting device installed in the substation.

[0040] Optionally, the device further includes a training unit for training the fire identification model, wherein the training unit, when training the fire identification model, is specifically used for:

[0041] Multiple training samples are obtained; each training sample includes a sample monitoring video, the static image features and dynamic image features corresponding to the sample monitoring video, and the pre-labeled real fire identification result, which indicates whether a fire has occurred or not.

[0042] The static and dynamic image features of each training sample are processed using the neural network model to be trained to obtain the predicted fire identification result corresponding to each training sample; wherein, the predicted fire identification result is either a fire has occurred or no fire has occurred.

[0043] The model loss of the neural network model to be trained is determined based on the deviation between the actual fire identification result and the predicted fire identification result for each sample.

[0044] Determine whether the model loss meets the preset convergence condition;

[0045] If the model loss does not meet the convergence condition, update the parameters of the neural network model to be trained according to the model loss, and return to the step of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample, until the model loss meets the convergence condition.

[0046] If the model loss satisfies the convergence condition, the neural network model to be trained is determined as a fire identification model.

[0047] A third aspect of this application provides an online fire monitoring system for substations, comprising:

[0048] A video acquisition module consisting of multiple camera devices installed in the substation is used to capture the first monitoring video of the substation;

[0049] The video analysis module is used to execute the substation online fire monitoring method provided in any one of the first aspects of this application based on the first monitoring video captured by the video acquisition module.

[0050] A video management module is used to store the first monitoring video and display the first monitoring video on at least one display.

[0051] The linkage fire suppression module includes at least an early warning device for outputting fire early warning information based on the control of the video analysis module.

[0052] The fourth aspect of this application provides a computer storage medium for storing a computer program, which, when executed, is specifically used to implement the substation online fire monitoring method provided in any one of the first aspects of this application.

[0053] This application provides a method, device, system, and storage medium for online fire monitoring of substations. The method includes: obtaining a first monitoring video of the substation; performing video preprocessing on the first monitoring video to obtain a second monitoring video; wherein the video preprocessing includes image cropping and image filtering; determining whether the brightness of an image frame in the second monitoring video is greater than or equal to a preset brightness threshold; when the brightness of an image frame is greater than or equal to the brightness threshold, reducing the background brightness of each image frame in the second monitoring video to obtain a third monitoring video; extracting static image features representing the presence or absence of smoke and flames from the second monitoring video, and extracting dynamic image features representing the amplitude of light change from the third monitoring video; processing the static image features and dynamic image features using a pre-trained fire identification model to obtain a fire identification result; wherein the fire identification model includes a static sub-model for processing static image features and a dynamic sub-model for processing dynamic images; when the fire identification result indicates that a fire has occurred, controlling an early warning device to output fire early warning information. This solution uses a fire identification model to analyze real-time photos of substation equipment to detect whether a fire has broken out in the substation. Compared with the physical quantities monitored by fire detectors, the photos taken by the camera equipment are almost unaffected by the outdoor environment, so this solution has higher accuracy. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0055] Figure 1 A schematic diagram of the structure of an online fire monitoring system for a substation provided in this application embodiment;

[0056] Figure 2 An example photograph of a substation fire provided as an embodiment of this application;

[0057] Figure 3 This application provides a superpixel-marked image of the fire area in a substation during a fire, as shown in an embodiment of the present application.

[0058] Figure 4 A flowchart illustrating an online fire monitoring method for substations provided in this application embodiment;

[0059] Figure 5 This is a schematic diagram of an online fire monitoring device for a substation, provided as an embodiment of this application. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Please see Figure 1 This is a schematic diagram of the structure of an online fire monitoring system for a substation provided in an embodiment of this application. The system may include a video acquisition module 101, a video analysis module 102, a video management module 103, and a linkage fire extinguishing module 104.

[0062] The video capture module includes multiple camera devices, such as Figure 1 The camera devices 1 to n, such as cameras, monitors, etc., can be installed in the substations that require fire monitoring to capture real-time video of the substation. This video is recorded as the first monitoring video.

[0063] The first monitoring video captured by the video acquisition module can be transmitted to the video analysis module.

[0064] The video analysis module has functions such as image preprocessing, image feature extraction, and smoke and flame recognition. In this embodiment, after the video analysis module obtains the first monitoring video, it can execute the online fire monitoring method for substations provided in this application based on the first monitoring video, thereby obtaining the fire identification result of the monitored substation.

[0065] Furthermore, when the fire identification result indicates that a fire has occurred in the substation, the video analysis module can send corresponding control commands to the linkage fire extinguishing module, thereby controlling the linkage fire extinguishing module to perform early warning and active fire extinguishing measures.

[0066] The video management module includes a hard disk recorder, a video distributor, a video matrix, and at least one display, wherein the display may be a color display.

[0067] The hard disk recorder is used to store the captured first monitoring video, and the video distributor is used to distribute the first monitoring video to at least one display, so that the first monitoring video can be played on these displays in real time.

[0068] The video matrix consists of multiple displays arranged in a matrix. The video matrix can simultaneously display the first monitoring video and the image of the fire area obtained after processing the first monitoring video based on the fire identification results on multiple displays for operation and maintenance personnel to view.

[0069] For example, please see Figure 2 This is a frame from the first monitoring video captured when the fire occurred. By analyzing this first monitoring video, the video analysis module can determine that a fire has occurred at the substation. Figure 2 In the image shown, the burned area during the fire was marked using superpixel labeling. The marked image is as follows: Figure 3 As shown, the video matrix can be displayed on multiple screens. Figure 2 The real-life scene shown, and Figure 3 The image shown marks the area on fire, making it easier for substation maintenance personnel to analyze the current fire situation.

[0070] The fire suppression linkage module can include early warning equipment and fire-fighting equipment, with the fire-fighting equipment located in the substation. When the video analysis module determines that a fire has occurred, it can control the early warning equipment in the fire suppression linkage module to output fire warning information, such as ringing a bell or automatically dialing the fire alarm number. At the same time, it can control the fire-fighting equipment in the fire suppression linkage module to actively extinguish the fire.

[0071] according to Figure 1 The substation online fire monitoring system shown in this application provides a substation online fire monitoring method. Please refer to [link to relevant documentation]. Figure 4 Here is a flowchart of the method, which may include the following steps:

[0072] The first aspect of this application provides a method for online fire monitoring of a substation, including:

[0073] S401, obtained the first monitoring video of the substation.

[0074] The first monitoring video is... Figure 1 In the system shown, the video acquired by the video acquisition module can have a duration equal to a preset monitoring duration. The monitoring duration can be preset according to actual conditions and is not limited. For example, the monitoring duration can be set to 1 minute.

[0075] Taking a monitoring duration of 1 minute as an example, the video acquisition module can capture video of the substation site in real time. Every minute of video capture, the video captured within that minute is sent to the video analysis module as a first monitoring video, so that the video analysis module can execute the method of this embodiment based on the first monitoring video.

[0076] S402, perform video preprocessing on the first monitoring video to obtain the second monitoring video.

[0077] Video preprocessing includes image cropping and image filtering.

[0078] Image cropping refers to cropping each image frame in the first monitoring video to a pre-defined target size. The advantage of image cropping is that by reducing the size of image frames, the amount of data that the subsequent fire detection network needs to process is reduced, thus obtaining fire detection results faster and improving the efficiency of the solution.

[0079] Image filtering refers to using specific filtering algorithms to remove noise data from each image frame of the first monitoring video. Image filtering can be implemented using any existing filtering algorithm, which will not be elaborated further.

[0080] The video, consisting of image frames that have been cropped and filtered, is the second monitoring video of S402.

[0081] S403, determine whether the brightness of the image frame in the second monitoring video is greater than or equal to the preset brightness threshold.

[0082] When the brightness of the image frame in the second monitoring video is greater than or equal to the brightness threshold, step S404 is executed; when the brightness of the image frame in the second monitoring video is not greater than or equal to, i.e. less than the brightness threshold, step S406 is executed.

[0083] The brightness of an image frame can be defined as the average brightness value of all pixels in that image frame.

[0084] Step S403 specifically involves determining the brightness of multiple image frames in the second monitoring video. For example, K image frames can be randomly selected from the second monitoring video, and it can be determined whether the brightness of these K image frames is greater than or equal to a brightness threshold. If they are all greater than or equal to the brightness threshold, then it is determined that the brightness of the image frames in the second monitoring video is greater than or equal to the brightness threshold, and step S404 is executed. If the brightness of at least one of the image frames is not greater than or equal to the brightness threshold, then it is determined that the brightness of the image frames in the second monitoring video is not greater than or equal to the brightness threshold, and step S406 is executed. K can be set as needed and is not limited. For example, K can be set to 2.

[0085] The aforementioned brightness threshold can be set as the average brightness of photos of a substation taken during a day with good lighting conditions (e.g., a sunny or cloudy day). Therefore, if the brightness of an image frame in the second monitoring video is greater than or equal to the brightness threshold, the first monitoring video can be considered as a monitoring video of a substation taken during a day with good lighting conditions. Conversely, if the brightness of an image frame in the second monitoring video is less than the brightness threshold, the first monitoring video can be considered as a monitoring video of a substation taken under poor lighting conditions, such as at night or during a rainy day.

[0086] S404, the background brightness of each image frame in the second monitoring video is reduced to obtain the third monitoring video.

[0087] Optionally, the specific execution process of step S404 may include:

[0088] Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video.

[0089] Background pixels refer to the pixels that belong to the background portion of an image frame.

[0090] Specifically, considering that substations are usually located in suburban areas, the environment around the substation can be photographed in advance during the day when lighting conditions are good, and the photographs can be used as background images. Based on this, when executing S404, the pixels in the image frames of the second monitoring video can be matched with the background image. If the pixels in a certain area of ​​the image frame of the second monitoring video have a high similarity to the background image, then the pixels in that area can be considered to belong to the background area of ​​the image frame, that is, the pixels in that area are determined to be background pixels.

[0091] The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

[0092] The advantage of executing steps S403 and S404 is that when the first monitoring video is captured during the day with good lighting conditions, even if a substation catches fire, the brightness of the flames will be masked by the strong ambient light. This prevents the subsequent fire identification model from determining whether the substation is on fire by processing dynamic image features, negatively impacting the final fire identification result. By reducing the background brightness, the brightness of the ambient light in the video captured during the day with good lighting conditions can be weakened, thereby highlighting the brightness of any potential flames. This allows the method in this embodiment to obtain accurate fire identification results even in daytime scenes with good lighting conditions by analyzing static and dynamic image features through the fire identification model.

[0093] S405, extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and extract dynamic image features representing the amplitude of light change from the third monitoring video.

[0094] The static image features can specifically be composed of multiple static feature matrices, each corresponding to an image frame in the second monitoring video. During step S405, several image frames can be extracted from the second monitoring video at certain intervals, for example, one frame can be extracted every 10 image frames. Then, the static feature matrix of each extracted image frame is extracted. The set of these static feature matrices constitutes the static image features described in step S405.

[0095] The process of extracting a static feature matrix from a specific image frame can be as follows:

[0096] First, the image frame is divided into several pooling regions with a preset pooling size. For example, the pooling size can be set to 5*5. Based on this, the image frame can be divided into several 5*5 pooling regions. Each pooling region includes 5 rows and 5 columns of pixels in the image frame.

[0097] Then, for each pooling region, the average pixel value of all pixels in that pooling region is calculated, and the calculation result is determined as the feature element corresponding to that pooling region. Finally, the position of each pooling region and the feature elements corresponding to each pooling region are arranged into a matrix to obtain the static feature matrix of the image frame.

[0098] Understandably, the pooling size can also be set to other values ​​without limitation.

[0099] The dynamic image features can be composed of the brightness features of several image frames in the third monitoring video. When extracting dynamic image features, the same method as when extracting static image features can be used to extract image frames from the third monitoring video. Then, the brightness features of each extracted image frame can be extracted. The set of these brightness features is the dynamic image feature of the third monitoring video.

[0100] For a specific image frame, the process of extracting its brightness features can be as follows:

[0101] First, the image frame is divided into several pooling regions with preset pooling sizes, and the average brightness of all pixels within each pooling region is calculated as the average brightness of that region. Then, N pooling regions are selected from the image frame in descending order of average brightness. The average brightness of these pooling regions is arranged in descending order to form a vector, which represents the brightness feature of the image frame. Here, N is a preset positive integer, and its value can be set according to actual conditions without limitation.

[0102] S406 Extract static image features representing the presence or absence of smoke and flames, and dynamic image features representing the magnitude of changes in illumination from the second monitoring video.

[0103] The specific execution process of step S406 is the same as that of step S405. It is only necessary to replace the third monitoring video obtained by adjusting the brightness of the object from S405 with the second monitoring video. It will not be described again.

[0104] S407 uses a pre-trained fire identification model to process static and dynamic image features to obtain fire identification results.

[0105] The fire identification model includes a static sub-model for processing static image features and a dynamic sub-model for processing dynamic images.

[0106] When the fire identification result indicates that a fire has occurred, step S408 is executed; when the fire identification result indicates that no fire has occurred, this embodiment ends.

[0107] In step S407, the dynamic sub-model can process the input dynamic image features to obtain the illumination change amplitude of several bright areas in the substation during the capture of the first monitoring video. The static sub-model can process the input static image features to identify whether there are flames and smoke in each image frame of the first monitoring video.

[0108] If the dynamic sub-model's illumination change exceeds a set threshold, and the static sub-model identifies flames and smoke in at least one image frame of the first monitoring video, then the fire identification result can be determined as a fire occurring; otherwise, the fire identification result can be determined as no fire occurring.

[0109] The aforementioned threshold for the change range can be set based on the brightness changes when equipment catches fire in an actual substation.

[0110] In some optional embodiments, when the fire identification result indicates that a fire has occurred, the fire identification model can also output fire area information to indicate the area on fire in the video. For example, this fire area information can be used... Figure 3 The superpixel markers shown are displayed in the image frames of the first monitoring video, or they can be displayed in the image frames of the first monitoring video using a curved frame.

[0111] S408 controls the output of fire warning information by the early warning equipment.

[0112] Understandably, in real-world application scenarios, Figure 1The substation online fire monitoring system shown can capture the first monitoring video of the substation in real time. Each time a segment of the first monitoring video is captured, the monitoring method described in the above embodiment is executed once based on the first monitoring video to obtain the corresponding fire identification result, thereby realizing all-weather substation fire monitoring.

[0113] Fire warning information can take many forms. For example, fire warning information may include ringing bells, fire broadcasts, etc.

[0114] Optionally, after the control and early warning equipment outputs fire early warning information, it also includes:

[0115] Activate at least one fire-fighting device installed in the substation.

[0116] In some optional embodiments, since there are multiple fire-fighting devices distributed in different locations within the substation, in the above steps, the location of the fire in the substation can be determined first based on the fire area information output by the fire identification model. Then, the nearest fire-fighting devices can be identified and activated to extinguish the fire.

[0117] Optionally, in addition to outputting fire warning information and activating fire-fighting equipment, the video analysis module can also control the linkage fire extinguishing module to automatically dial the fire alarm number when a fire is confirmed.

[0118] Optionally, the process of training a fire identification model includes:

[0119] A1, obtain multiple training samples.

[0120] Each training sample includes a sample monitoring video, the corresponding static image features and dynamic image features, and the pre-labeled real fire identification results, which indicate whether a fire has occurred or not.

[0121] The training samples can be selected and processed from multiple monitoring videos of substations taken within a recent period (e.g., the last month).

[0122] The filtering process involves removing videos from multiple monitoring videos that have excessively low resolution, such as those below a preset resolution threshold. The advantage of this is that videos with excessively low resolution are difficult to analyze accurately to determine if a fire has occurred, and including these videos in the sample reduces training efficiency. Filtering out these videos, however, improves training efficiency.

[0123] The processing of sample monitoring videos can include two steps: cropping and normalization.

[0124] Cropping refers to removing irrelevant areas from a monitoring video captured from multiple angles, retaining only the target areas relevant to fire monitoring. Specifically, fire monitoring in substations primarily monitors whether various electrical devices within the substation are on fire. Therefore, cropping can remove areas that do not contain electrical equipment, retaining only the target areas within the video frames that do contain the electrical equipment.

[0125] Normalization refers to converting a large number of videos of different sizes after screening and cropping into a uniform sample size, so that the video image resources are converted into a standard and uniform format, thereby ensuring the correctness of the input samples and improving the accuracy of the results.

[0126] Through the above processing, multiple sample monitoring videos can be obtained from multiple monitoring videos of substations captured in recent times.

[0127] After obtaining these sample monitoring videos, the actual fire identification result of each sample monitoring video can be determined by manually viewing and labeling them. Specifically, if an electrical device in a sample monitoring video is on fire, the actual fire result of the sample monitoring video is labeled as a fire. If an electrical device in a sample monitoring video is not on fire, the actual fire result of the sample monitoring video is labeled as no fire.

[0128] On the other hand, the static image features and dynamic image features of each sample monitoring video can be extracted by following the process of extracting static image features and dynamic image features in step S405 mentioned above.

[0129] Ultimately, a training sample can be formed by combining a sample monitoring video, the corresponding real fire identification result, static image features, and dynamic image features.

[0130] A2 uses the neural network model to be trained to process the static and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample.

[0131] Among them, the predicted fire identification result is either a fire has occurred or no fire has occurred.

[0132] The model structure of the neural network model to be trained can be any existing model structure, without any restrictions.

[0133] Before executing step A2 for the first time, the values ​​of each parameter in the neural network model to be trained can be randomly set. This process is called the initialization of the neural network model to be trained.

[0134] The specific execution process of step A2 is the same as that of step S407, and will not be repeated here.

[0135] A3. Based on the deviation between the actual fire identification result and the predicted fire identification result for each sample, determine the model loss of the neural network model to be trained.

[0136] One way to determine the model loss is:

[0137] The actual fire identification result and the predicted fire identification result of each training sample are compared one by one to see if they are consistent. Then, the proportion of training samples with inconsistent results is counted out of all training samples, and this proportion is determined as the model loss of the neural network model to be trained.

[0138] A4. Determine whether the model loss meets the preset convergence conditions.

[0139] The convergence condition can be set as follows: the model loss is less than a preset convergence loss threshold, or the number of training iterations is greater than or equal to a preset maximum number of training iterations, where the number of training iterations can be defined as the number of times step A5 is executed.

[0140] A5. If the model loss does not meet the convergence condition, update the parameters of the neural network model to be trained according to the model loss, and return to the step of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample, until the model loss meets the convergence condition.

[0141] In step A5, the gradient backpropagation algorithm can be used to calculate the update amount of each parameter in the neural network model to be trained based on the model loss, and then the value of each parameter in each neural network model to be trained is updated according to the update amount.

[0142] The specific execution process of the gradient backpropagation algorithm can be found in relevant existing technologies, and will not be elaborated further.

[0143] A6. If the model loss satisfies the convergence condition, the neural network model to be trained is determined as the fire identification model.

[0144] This application provides an online fire monitoring method for substations. The method includes: preprocessing a first monitoring video of the substation to obtain a second monitoring video; when the brightness of an image frame in the second monitoring video is greater than or equal to a brightness threshold, reducing the background brightness of that image frame to obtain a third monitoring video; extracting static image features representing the presence or absence of smoke and flames from the second monitoring video, and extracting dynamic image features representing the amplitude of light change from the third monitoring video; and processing the static and dynamic image features using a fire identification model to obtain a fire identification result. This solution uses a fire identification model to analyze real-time photographs of substation equipment to detect whether a fire has occurred in the substation. Compared with the physical quantities monitored by fire detectors, the photographs taken by the camera equipment are almost unaffected by the outdoor environment, therefore this solution has higher accuracy.

[0145] According to the online fire monitoring method for substations provided in the embodiments of this application, the embodiments of this application also provide an online fire monitoring device for substations. Please refer to [link to relevant documentation]. Figure 5 This is a schematic diagram of the structure of the device, which may include the following units.

[0146] The online fire monitoring device for substations provided in this embodiment can be considered as... Figure 1 The video analysis module in the substation online fire monitoring system shown is an example.

[0147] Unit 501 is used to acquire the first monitoring video of the substation;

[0148] The preprocessing unit 502 is used to perform video preprocessing on the first monitoring video to obtain the second monitoring video; wherein, the video preprocessing includes image cropping and image filtering;

[0149] The judgment unit 503 is used to determine whether the brightness of the image frame in the second monitoring video is greater than or equal to a preset brightness threshold.

[0150] The adjustment unit 504 is used to reduce the background brightness of each image frame in the second monitoring video when the brightness of the image frame is greater than or equal to the brightness threshold, so as to obtain the third monitoring video.

[0151] Extraction unit 505 is used to extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and to extract dynamic image features representing the amplitude of light change from the third monitoring video.

[0152] The identification unit 506 is used to process static image features and dynamic image features using a pre-trained fire identification model to obtain fire identification results; wherein, the fire identification model includes a static sub-model for processing static image features and a dynamic sub-model for processing dynamic images.

[0153] The execution unit 507 is used to control the early warning device to output fire early warning information when the fire identification result indicates that a fire has occurred.

[0154] Optionally, when adjusting the background brightness of each image frame in the second monitoring video to obtain the third monitoring video, the adjustment unit 504 is specifically used for:

[0155] Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video.

[0156] The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

[0157] Optionally, after the execution unit 507 controls the early warning device to output fire early warning information, it is also used for:

[0158] Activate at least one fire-fighting device installed in the substation.

[0159] Optionally, the device also includes a training unit 508 for training the fire recognition model. When training the fire recognition model, the training unit specifically performs the following functions:

[0160] Multiple training samples are obtained; each training sample includes a sample monitoring video, the static image features and dynamic image features corresponding to the sample monitoring video, and the pre-labeled real fire identification results, which indicate whether a fire has occurred or not.

[0161] The static and dynamic image features of each training sample are processed using the neural network model to be trained to obtain the predicted fire identification result for each training sample; wherein the predicted fire identification result is either a fire has occurred or no fire has occurred.

[0162] The model loss of the neural network model to be trained is determined based on the deviation between the actual fire identification result and the predicted fire identification result for each sample.

[0163] Determine whether the model loss meets the preset convergence conditions;

[0164] If the model loss does not meet the convergence condition, update the parameters of the neural network model to be trained according to the model loss, and return to the step of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample, until the model loss meets the convergence condition.

[0165] If the model loss satisfies the convergence condition, the neural network model to be trained is determined as the fire identification model.

[0166] The specific working principle and beneficial effects of the substation online fire monitoring device provided in this embodiment can be found in the relevant steps and beneficial effects of the substation online fire monitoring method provided in this application embodiment, and will not be repeated here.

[0167] This application also provides a computer storage medium for storing a computer program. When the stored computer program is executed, it is specifically used to implement the substation online fire monitoring method provided in this application.

[0168] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0169] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0170] Those skilled in the art will be able to implement or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for online fire monitoring of substations, characterized in that, include: Obtain the first monitoring video of the substation; The first monitoring video is preprocessed to obtain the second monitoring video; wherein, the video preprocessing includes image cropping and image filtering; Determine whether the brightness of the image frame in the second monitoring video is greater than or equal to a preset brightness threshold; When the brightness of the image frame is greater than or equal to the brightness threshold, the background brightness of each image frame in the second monitoring video is reduced to obtain the third monitoring video; Extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and extract dynamic image features representing the amplitude of light change from the third monitoring video; The static image features and the dynamic image features are processed using a pre-trained fire identification model to obtain fire identification results; wherein, the fire identification model includes a static sub-model for processing the static image features and a dynamic sub-model for processing the dynamic image features; When the fire identification result indicates that a fire has occurred, the control and early warning equipment outputs fire early warning information.

2. The method according to claim 1, characterized in that, The step of reducing the background brightness of each image frame in the second monitoring video to obtain the third monitoring video includes: Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video. The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

3. The method according to claim 1, characterized in that, After the control and early warning device outputs fire early warning information, it also includes: Activate at least one fire-fighting device installed in the substation.

4. The method according to claim 1, characterized in that, The process of training the fire identification model includes: Multiple training samples are obtained; each training sample includes a sample monitoring video, the static image features and dynamic image features corresponding to the sample monitoring video, and the pre-labeled real fire identification result, which indicates whether a fire has occurred or not. The static and dynamic image features of each training sample are processed using the neural network model to be trained to obtain the predicted fire identification result corresponding to each training sample; wherein, the predicted fire identification result is either a fire has occurred or no fire has occurred; The model loss of the neural network model to be trained is determined based on the deviation between the actual fire identification result and the predicted fire identification result for each sample. Determine whether the model loss meets the preset convergence condition; If the model loss does not meet the convergence condition, the parameters of the neural network model to be trained are updated according to the model loss, and the process of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample is returned until the model loss meets the convergence condition. If the model loss satisfies the convergence condition, the neural network model to be trained is determined as a fire identification model.

5. An online fire monitoring device for substations, characterized in that, include: The acquisition unit is used to acquire the first monitoring video of the substation. A preprocessing unit is used to perform video preprocessing on the first monitoring video to obtain a second monitoring video; wherein, the video preprocessing includes image cropping and image filtering; The judgment unit is used to determine whether the brightness of the image frame in the second monitoring video is greater than or equal to a preset brightness threshold. An adjustment unit is used to reduce the background brightness of each image frame in the second monitoring video when the brightness of the image frame is greater than or equal to the brightness threshold, so as to obtain a third monitoring video. The extraction unit is used to extract static image features representing the presence or absence of smoke and flames from the second monitoring video, and to extract dynamic image features representing the amplitude of light change from the third monitoring video. The identification unit is used to process the static image features and the dynamic image features using a pre-trained fire identification model to obtain a fire identification result; wherein, the fire identification model includes a static sub-model for processing the static image features and a dynamic sub-model for processing the dynamic image features; The execution unit is used to control the early warning device to output fire early warning information when the fire identification result indicates that a fire has occurred.

6. The apparatus according to claim 5, characterized in that, When the adjustment unit lowers the background brightness of each image frame in the second monitoring video to obtain the third monitoring video, it is specifically used for: Each image frame in the second monitoring video is matched with a pre-acquired background image to determine the background pixels of each image frame in the second monitoring video. The brightness of the background pixels in each image frame of the second monitoring video is reduced to the lower limit of the brightness value range to obtain the third monitoring video.

7. The apparatus according to claim 5, characterized in that, After the execution unit controls the early warning device to output fire early warning information, it is also used for: Activate at least one fire-fighting device installed in the substation.

8. The apparatus according to claim 5, characterized in that, The device further includes a training unit for training the fire identification model, wherein the training unit, when training the fire identification model, is specifically used for: Multiple training samples are obtained; each training sample includes a sample monitoring video, the static image features and dynamic image features corresponding to the sample monitoring video, and the pre-labeled real fire identification result, which indicates whether a fire has occurred or not. The static and dynamic image features of each training sample are processed using the neural network model to be trained to obtain the predicted fire identification result corresponding to each training sample; wherein, the predicted fire identification result is either a fire has occurred or no fire has occurred; The model loss of the neural network model to be trained is determined based on the deviation between the actual fire identification result and the predicted fire identification result for each sample. Determine whether the model loss meets the preset convergence condition; If the model loss does not meet the convergence condition, the parameters of the neural network model to be trained are updated according to the model loss, and the process of using the neural network model to be trained to process the static image features and dynamic image features of each training sample to obtain the predicted fire identification result corresponding to each training sample is returned until the model loss meets the convergence condition. If the model loss satisfies the convergence condition, the neural network model to be trained is determined as a fire identification model.

9. A substation online fire monitoring system, characterized in that, include: A video acquisition module consisting of multiple camera devices installed in the substation is used to capture the first monitoring video of the substation; The video analysis module is used to execute the substation online fire monitoring method as described in any one of claims 1 to 4 based on the first monitoring video captured by the video acquisition module. A video management module is used to store the first monitoring video and display the first monitoring video on at least one monitor; The linkage fire suppression module includes at least an early warning device for outputting fire early warning information based on the control of the video analysis module.

10. A computer storage medium, characterized in that, Used to store computer programs, which, when executed, are specifically used to implement the substation online fire monitoring method as described in any one of claims 1 to 4.

Images