Forest fire identification method and device for power transmission corridor, electronic equipment and storage medium
By using smoke detection models and multimodal models to process visible light images in transmission corridors, the problem of wildfire identification in transmission corridors was solved, accurate identification of smoke areas and timely judgment of wildfires were achieved, ensuring the safety of the power system.
Patent Information
- Application Number
- CN202510760151.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-09-09
AI Technical Summary
Existing technologies make it difficult to effectively identify wildfires in transmission corridors, causing fires to pose a threat to the safe operation of the power system. Failure to detect and deal with them in a timely manner may lead to serious accidents such as broken wires and collapsed transmission towers.
By acquiring visible light images of the transmission corridor, a smoke detection model based on a deep learning algorithm is used to generate a segmentation mask of the smoke area. This is then combined with a multimodal model for further judgment to identify the presence of wildfires, including data augmentation processing and training sample optimization to improve the robustness and accuracy of the model.
It has achieved accurate identification of smoke areas in transmission corridors, can effectively distinguish wildfire smoke from natural phenomena, improves the accuracy and timeliness of wildfire identification, and ensures the safe and stable operation of the power grid.
Smart Images
Figure CN120612609A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wildfire identification, and in particular to a method, device, electronic equipment and storage medium for identifying wildfires in a power transmission corridor. Background Art
[0002] A transmission corridor refers to a specific area along a high-voltage transmission line, encompassing transmission towers, conductors, and their surroundings. As crucial routes for power transmission, transmission corridors often traverse natural areas such as mountains, forests, and grasslands. These areas are characterized by complex terrain and lush vegetation, making them highly susceptible to natural disasters. Especially during droughts or extreme weather conditions, vegetation becomes more flammable. Once a fire breaks out, it can spread rapidly, posing a serious threat to the safe operation of the transmission corridor. Wildfires are a major safety hazard along transmission corridors. The high temperatures generated by fires can directly impact the physical properties of transmission lines. For example, high temperatures can cause thermal expansion of conductors, leading to abnormal sag and increasing the risk of loss of contact between conductors and vegetation or the ground. Furthermore, the large amounts of smoke and combustion particles generated by wildfires can adhere to transmission lines and insulators, causing flashover and other faults, further impacting the normal operation of transmission lines. Failure to promptly detect and address wildfires can lead to conductor breakage, tower collapse, and even widespread power outages, seriously threatening the stability of the power system and regional power supply security. Therefore, identifying wildfires within the transmission corridor area can ensure the safe and stable operation of the power grid. Summary of the Invention
[0003] The present invention provides a method, device, electronic device, and storage medium for identifying wildfires in power transmission corridors. By implementing the present invention, wildfires in power transmission corridors can be effectively identified.
[0004] An embodiment of the present invention provides a method for identifying wildfires in a power transmission corridor, comprising:
[0005] Obtain visible light images of the transmission corridor to be studied;
[0006] Inputting the visible light image into a preset smoke detection model so that the smoke detection model generates a segmentation mask of the smoke region in the visible light image based on the visible light image; wherein the smoke detection model is trained using a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke region in the visible light image sample;
[0007] Annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image;
[0008] Obtaining a first question from a user; the first question includes: whether there is a wildfire and how to answer the question;
[0009] The first question information and the smoke-annotated image are input into a preset multimodal model, so that the multimodal model generates a determination result on whether a wildfire exists in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image in the form of an answer to the question in the first question information; wherein the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether a wildfire exists.
[0010] Furthermore, the smoke detection model is trained in the following way:
[0011] Obtaining a number of first training samples;
[0012] Each first training sample is input into the smoke detection model in sequence, and the smoke detection model is trained until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; based on the segmentation mask of the predicted smoke area and the corresponding semantic label, a first loss function value is calculated; and the smoke detection model is updated according to the first loss function value.
[0013] Furthermore, the multimodal model is trained in the following way:
[0014] Obtaining a number of second training samples;
[0015] Each second training sample is input into the multimodal model in sequence, and the multimodal model is trained until a preset second training number is reached; wherein, each time the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label, the second loss function value is calculated; and the multimodal model is updated according to the second loss function value.
[0016] Furthermore, before inputting each first training sample into the smoke detection model in sequence and training the smoke detection model until a preset first training number is reached, the method further includes:
[0017] For the visible light image samples in the first training samples of each batch, data augmentation processing is performed to generate updated visible light image samples; wherein the data augmentation processing includes any one or a combination of random flipping, random scaling, random translation, random rotation, random adjustment of image brightness, random addition of Gaussian noise and random cropping.
[0018] Furthermore, after inputting the first question information and the smoke-annotated image into a preset multimodal model, so that the multimodal model generates a determination result of whether a wildfire exists in the transmission corridor to be studied in the form of an answer to the question in the first question information based on the user's first question information and the smoke-annotated image, the method further includes:
[0019] Send an alarm message when there is a wildfire in the transmission corridor to be studied.
[0020] Based on the above method embodiments, the present invention provides corresponding device embodiments.
[0021] An embodiment of the present invention provides a wildfire identification device for a power transmission corridor, comprising: an image acquisition module, a smoke detection module, a smoke labeling module, and a wildfire identification module;
[0022] The image acquisition module is used to acquire visible light images of the transmission corridor to be studied;
[0023] The smoke detection module is configured to input the visible light image into a preset smoke detection model, so that the smoke detection model generates a segmentation mask of the smoke region in the visible light image based on the visible light image; wherein the smoke detection model is trained using a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke region in the visible light image sample;
[0024] The smoke annotation module is used to annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image;
[0025] The wildfire identification module is used to obtain a user's first question information; the first question information includes: whether a wildfire exists and a method for answering the question; the first question information and the smoke-annotated image are input into a preset multimodal model, so that the multimodal model generates a determination result on whether a wildfire exists in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image, in the method for answering the question in the first question information; wherein, the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether a wildfire exists.
[0026] Furthermore, the wildfire identification device for the power transmission corridor further includes: a smoke detection model training module;
[0027] The smoke detection model training module is used to obtain a number of first training samples; input each first training sample into the smoke detection model in sequence, and train the smoke detection model until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; calculates a first loss function value based on the segmentation mask of the predicted smoke area and the corresponding semantic label; and updates the smoke detection model based on the first loss function value.
[0028] Furthermore, the wildfire identification device for the power transmission corridor further includes: a multimodal model training module;
[0029] The multimodal model training module is used to obtain a number of second training samples; input each second training sample into the multimodal model in sequence, and train the multimodal model until a preset second training number is reached; wherein, when the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; calculates the second loss function value based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label; and updates the multimodal model based on the second loss function value.
[0030] Based on the above method embodiment, the present invention provides a corresponding electronic device embodiment.
[0031] An embodiment of the present invention provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying wildfires in the transmission corridor described in any one of the above-mentioned method embodiments.
[0032] Based on the above method embodiment, the present invention provides a corresponding storage medium embodiment.
[0033] An embodiment of the present invention provides a storage medium having a computer program stored thereon, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the method for identifying wildfires in a transmission corridor as described in any one of the above-mentioned method embodiments.
[0034] Compared with the prior art, the present invention has the following beneficial effects:
[0035] The embodiments of the present invention provide a method, device, electronic device, and storage medium for identifying wildfires in power transmission corridors. The method obtains a visible light image of the power transmission corridor to be studied; inputs the visible light image into a preset smoke detection model, generates a segmentation mask of the smoke area in the visible light image, and further inputs the segmentation mask of the smoke area into a preset multimodal model to further determine and identify the wildfire. The smoke detection model can identify the smoke area. On this basis, by further using the multimodal model to further determine the smoke area, it can effectively exclude natural phenomena such as clouds and haze that have similar visual characteristics to wildfire smoke, and effectively identify the occurrence of wildfires in the power transmission corridor. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 The figure is a flow chart of a method for identifying wildfires in a power transmission corridor provided by one embodiment of the present invention.
[0037] Figure 2 This is a structural diagram of a wildfire identification device for a power transmission corridor provided by one embodiment of the present invention. DETAILED DESCRIPTION
[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0039] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying wildfires in a power transmission corridor, comprising at least the following steps:
[0040] Step S1: Obtain a visible light image of the transmission corridor to be studied.
[0041] It is important to note that visible light images of the transmission corridor under study are obtained for subsequent smoke detection and wildfire identification. Visible light images can intuitively reflect the actual environment along the transmission corridor, including information such as transmission towers, conductors, and surrounding terrain and vegetation. By deploying fixed monitoring equipment or drone inspection equipment along the transmission corridor, visible light images can be collected regularly or in real time to ensure continuous monitoring of fire conditions in the transmission corridor. The obtained visible light images contain rich scene information and serve as the basic data source for subsequent smoke detection models and multimodal models for identification and analysis.
[0042] Step S2: Input the visible light image into a preset smoke detection model so that the smoke detection model generates a segmentation mask of the smoke area in the visible light image based on the visible light image; wherein the smoke detection model is trained by a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke area in the visible light image sample.
[0043] Specifically, the visible light image is input into a preset smoke detection model, which generates a segmentation mask for the smoke region within the visible light image. Based on a deep learning algorithm, the smoke detection model can accurately detect smoke regions in complex scenes and generate corresponding segmentation masks, precisely annotating the location and shape of smoke within the image. This segmentation mask effectively highlights smoke regions, providing reliable data support for subsequent wildfire identification.
[0044] The smoke detection model is trained using a number of first training samples, each of which includes a visible light image sample of a transmission corridor and a corresponding semantic label. The semantic label is a segmentation mask of the smoke region within the visible light image sample, annotating the specific location and boundary information of the smoke within the image. By collecting a large number of training samples under different weather conditions, lighting conditions, and complex environments, and performing detailed labeling, the smoke detection model has acquired the ability to identify smoke in multiple scenarios.
[0045] Leveraging extensive training data and optimized deep learning algorithms, the smoke detection model effectively adapts to the changing environment along power transmission corridors, enabling rapid and accurate segmentation of smoke areas. The generated segmentation mask not only captures the smoke's edge contours but also reflects its coverage and density, providing precise preliminary identification results for subsequent multimodal model analysis.
[0046] In a preferred embodiment, the smoke detection model is trained in the following manner:
[0047] Obtaining a number of first training samples;
[0048] Each first training sample is input into the smoke detection model in sequence, and the smoke detection model is trained until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; based on the segmentation mask of the predicted smoke area and the corresponding semantic label, a first loss function value is calculated; and the smoke detection model is updated according to the first loss function value.
[0049] Specifically, a number of first training samples are obtained as a training dataset for the smoke detection model. Each first training sample includes a visible light image sample of a transmission corridor and a corresponding semantic label. The semantic label is a segmentation mask of the smoke region in the visible light image sample, which is used to annotate the specific location and boundary information of the smoke in the image.
[0050] Each first training sample is sequentially input into the smoke detection model, and the smoke detection model is trained until a preset first number of training cycles is reached. Upon receiving each first training sample, the smoke detection model extracts and analyzes features of the image based on the current model parameters and outputs a segmentation mask for the predicted smoke region corresponding to the training sample. The predicted segmentation mask represents the model's determination of the smoke region, including the spatial location and morphological characteristics of the smoke.
[0051] The predicted smoke region segmentation mask is then compared with the actual semantic labels of the training samples to calculate the first loss function value. This loss function measures the difference between the model's predictions and the true labels. A smaller first loss function value indicates a higher prediction accuracy. During training, an optimization algorithm is used to update the smoke detection model's parameters based on the calculated first loss function value, gradually reducing the loss function value and improving the model's recognition capabilities.
[0052] Through multiple rounds of training and parameter optimization, the smoke detection model continuously learns and extracts characteristic information about smoke areas, enhancing its ability to detect smoke in various forms, concentrations, and background environments. Training is complete when the number of training cycles reaches the preset first training epoch or the loss function converges to the target range. A fully trained smoke detection model can accurately identify smoke areas in transmission corridors in real-world applications, providing reliable baseline data for subsequent wildfire identification.
[0053] In a preferred embodiment, before inputting each first training sample into the smoke detection model in sequence and training the smoke detection model until a preset first training number is reached, the method further includes:
[0054] For the visible light image samples in the first training samples of each batch, data augmentation processing is performed to generate updated visible light image samples; wherein the data augmentation processing includes any one or a combination of random flipping, random scaling, random translation, random rotation, random adjustment of image brightness, random addition of Gaussian noise and random cropping.
[0055] Specifically, data augmentation is performed on the visible light image samples in the first training sample of each batch to generate updated visible light image samples. Data augmentation is an effective image processing technique that aims to increase the diversity of training data by performing various random transformations on the original image samples, thereby improving the robustness and generalization ability of the smoke detection model, enabling it to have stronger recognition capabilities in complex environments.
[0056] Data augmentation processing includes any one or a combination of random flipping, random scaling, random translation, random rotation, random adjustment of image brightness, random addition of Gaussian noise, and random cropping. Among them, random flipping can mirror the image horizontally or vertically to simulate the perspective change in real scenes; random scaling can adjust the image to different degrees to adapt the model to smoke characteristics at different distances and scales; random translation simulates camera jitter or perspective offset, helping the model cope with slight displacement changes; random rotation rotates the image within a certain angle range to enhance the model's smoke recognition ability at different angles; random adjustment of image brightness increases or decreases the image brightness value to simulate different lighting conditions such as morning, dusk, or cloudy days; random addition of Gaussian noise simulates sensor noise or image interference in bad weather to improve the model's noise resistance performance; random cropping, by cropping and reconstructing the image at different ratios, makes the model focus on local features in the image, improving the ability to detect smoke in small areas.
[0057] Through this data augmentation process, the updated visible light image samples generated not only include the smoke information in the original scene, but also introduce diverse scene changes and noise interference, making the smoke detection model more robust and adaptable to different environments. During the training process, the data-augmented image samples are input into the smoke detection model along with the corresponding semantic labels, helping the model to accurately identify smoke areas even in the complex environments of real-world scenes.
[0058] Step S3: annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image.
[0059] Specifically, the visible light image is annotated based on the segmentation mask of the smoke region in the visible light image to generate a smoke annotated image. The segmentation mask clearly identifies the specific smoke region in the visible light image, including the smoke boundary, shape, and coverage. By combining the segmentation mask with the original visible light image, the smoke region can be intuitively highlighted, forming a clear smoke annotated image.
[0060] During the annotation process, you can use various visual identification methods to mark the smoke area. For example, you can use a semi-transparent color to cover the smoke area to clearly show the smoke's location and outline while preserving the original image information. Or, you can use outline tracing to mark the edge of the segmentation mask as a striking boundary line for easier observation and analysis. Furthermore, you can use different colors or shades of annotation to intuitively present the smoke's characteristics based on its concentration or distribution.
[0061] The generated smoke-annotated images not only provide intuitive input data for subsequent wildfire identification but also facilitate manual review and verification of the model's detection results. During training, the smoke-annotated images can also serve as input to the multimodal model, helping the model further analyze the characteristics of the smoke area and accurately determine whether the smoke is caused by a wildfire. In this way, smoke-annotated images effectively enhance the reliability and accuracy of wildfire identification in practical applications.
[0062] Step S4: Obtain the user's first question information; the first question information includes: whether there is a wildfire and how to answer the question.
[0063] Specifically, the user's first question is obtained. The first question includes the user's question about whether a wildfire exists in the current smoke-annotated image, along with the user's desired response. The user's first question is used to guide the multimodal model in analyzing and judging the smoke-annotated image, ensuring that the model's output meets the user's requirements.
[0064] In the first question, users can explicitly ask whether a wildfire exists in the target scene, such as "Is there a wildfire in the current image?" or "Is the detected smoke caused by a wildfire?" Furthermore, users can customize the specific wording of the question to suit different scenario requirements.
[0065] The first question also includes a response format, allowing users to choose from a variety of options based on their needs. This can range from a simple binary judgment, directly outputting a "yes" or "no" answer to quickly determine whether a wildfire exists; to a probabilistic score, outputting the model's confidence in the existence of a wildfire for further judgment; or, using natural language interpretation, generating a detailed analysis report from the model explaining the basis for the wildfire determination and relevant features.
[0066] By obtaining the user's first question information, the multimodal model can provide flexible and accurate judgment results based on user needs, meet the needs of wildfire identification in different scenarios, and provide more intelligent support for fire monitoring and early warning in the transmission corridor.
[0067] Step S5: Input the first question information and the smoke-annotated image into a preset multimodal model, so that the multimodal model generates a determination result of whether there is a wildfire in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image in the form of an answer to the question in the first question information; wherein the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether there is a wildfire.
[0068] Specifically, the first question and the annotated smoke image are input into a pre-set multimodal model. Based on the user's first question and the annotated smoke image, the multimodal model generates a determination result regarding whether a wildfire exists in the transmission corridor under investigation, according to the question-answering method specified in the first question. The multimodal model comprehensively analyzes the smoke characteristics in the annotated smoke image and, in light of the user's specific question requirements, provides a corresponding wildfire identification result.
[0069] The multimodal model is trained using several second training samples, each of which includes a smoke-annotated image sample, a user's second question, and a corresponding wildfire identification label. The smoke-annotated image sample is a visible light image with smoke-labeled areas, providing the spatial location and morphological characteristics of the smoke. The user's second question includes a specific question about the image and the expected response, guiding the model to generate targeted judgments in different scenarios. The wildfire identification label, based on the actual presence or absence of a wildfire, supervises the model's learning of wildfire characteristics in the image.
[0070] During training, the multimodal model continuously learns the correlation between smoke areas and wildfire characteristics through a large number of secondary training samples, improving the model's discriminative and reasoning capabilities. The model can not only identify typical wildfire smoke, but also accurately judge smoke in different forms, concentrations, and environmental conditions, effectively distinguishing wildfire smoke from similar smoke caused by natural phenomena such as clouds and haze.
[0071] Furthermore, the multimodal model's output is flexible and diverse, providing different types of judgment results based on user needs. If the user requests a simple binary judgment, the model can directly output a "yes" or "no" conclusion regarding the presence of a wildfire. If the user requires a more nuanced answer, the model can generate a confidence score indicating the probability of a wildfire. For scenarios requiring detailed analysis, the model can also output explanatory text based on image features to help users understand the basis for the judgment.
[0072] In this way, the multimodal model not only effectively improves the accuracy of wildfire identification in transmission corridors, but also meets the fire monitoring needs in different scenarios, providing power operation and maintenance personnel with fast and reliable decision support.
[0073] In a preferred embodiment, the multimodal model is trained in the following manner:
[0074] Obtaining a number of second training samples;
[0075] Each second training sample is input into the multimodal model in sequence, and the multimodal model is trained until a preset second training number is reached; wherein, each time the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label, the second loss function value is calculated; and the multimodal model is updated according to the second loss function value.
[0076] Specifically, several second training samples are obtained as a training dataset for the multimodal model. Each second training sample includes a smoke-annotated image sample, a second user question, and a corresponding wildfire identification label. The smoke-annotated image sample provides the spatial location and morphological characteristics of the smoke. The second user question specifies the specific question about the smoke and the desired response. The wildfire identification label indicates whether a wildfire is present in the image, providing supervision information for the model.
[0077] Each second training sample is sequentially input into the multimodal model, and the model is trained until the second set of training cycles is reached. During the training process, the multimodal model receives the smoke-annotated image sample and the user's second question. It then generates a predicted wildfire identification corresponding to the second training sample through multimodal fusion of image features and semantic information. The prediction result can be a binary judgment, a probability score, or a natural language interpretation, depending on the user's response to the question.
[0078] The model's predicted wildfire identification is then compared with the corresponding wildfire identification labels in the second training sample to calculate the second loss function value. This second loss function value measures the difference between the model's prediction and the true label. Different loss functions can be used for different types of prediction results. For example, for binary judgment results, the cross-entropy loss function can be used; for probability scoring results, the mean squared error loss function can be used; and for explanatory text generated by natural language, the cross-entropy loss function for sequence generation or the contrastive learning loss function can be used.
[0079] Based on the calculated second loss function value, an optimization algorithm was used to update the parameters of the multimodal model. By continuously optimizing the model parameters, the model gradually learned the deep relationship between smoke-annotated images and wildfire characteristics, enhancing its ability to identify wildfires in different scenarios. With increasing training times, the model's loss function value tended to converge, indicating a significant improvement in the model's recognition accuracy and stability.
[0080] After thorough training, the multimodal model can effectively adapt to complex transmission corridor scenarios and accurately identify wildfire characteristics in smoke-filled areas. Whether in sunny, cloudy, or foggy weather conditions, or in diverse terrain, lighting, and background environments, the multimodal model can generate accurate wildfire identification results, providing reliable technical support for practical wildfire monitoring and early warning.
[0081] In a preferred embodiment, after inputting the first question information and the smoke-annotated image into a preset multimodal model so that the multimodal model generates a determination result of whether a wildfire exists in the transmission corridor to be studied in the form of an answer to the question in the first question information based on the user's first question information and the smoke-annotated image, the method further includes:
[0082] Send an alarm message when there is a wildfire in the transmission corridor to be studied.
[0083] It's important to note that if a wildfire occurs along the transmission corridor under study, timely alerts will be sent so that relevant personnel can quickly take emergency measures. These alerts can include key information such as the fire's specific location, smoke coverage, and fire growth, helping operations and maintenance personnel and emergency management departments fully understand the fire situation and respond quickly.
[0084] When generating an alert, the multimodal model's results can be used to pinpoint the geographic coordinates of the identified wildfire area and annotate the fire location using map information. Furthermore, the alert can include a smoke annotated image and segmentation mask, visually demonstrating the specifics of the wildfire and facilitating further verification and analysis by relevant personnel.
[0085] To provide more comprehensive fire information, alarms can also include fire severity assessments. For example, by analyzing smoke density, color, and diffusion rate, fire intensity and spread can be inferred. Furthermore, historical meteorological data and environmental parameters such as current wind direction and speed can be used to predict the likely direction of a fire's spread, assisting in developing a scientific and rational firefighting plan.
[0086] Alarm information can be delivered through a variety of channels, including text messages, phone calls, emails, and mobile app notifications, ensuring that relevant personnel receive fire information as quickly as possible. For power dispatch centers, transmission operations and maintenance units, and emergency management departments, alarm information can also be pushed directly to a dedicated monitoring platform, enabling real-time monitoring and coordinated response to fire situations.
[0087] By sending warning information in a timely manner, the early warning and response efficiency of wildfires can be significantly improved, helping relevant personnel to intervene quickly in the early stages of the fire and take effective firefighting and protective measures, thereby minimizing the losses to transmission corridor equipment and the surrounding environment and ensuring the safe and stable operation of the power grid.
[0088] Based on the above method embodiments, the present invention provides corresponding device embodiments.
[0089] like Figure 2 As shown, an embodiment of the present invention provides a wildfire identification device for a power transmission corridor, comprising: an image acquisition module, a smoke detection module, a smoke marking module, and a wildfire identification module;
[0090] The image acquisition module is used to acquire visible light images of the transmission corridor to be studied;
[0091] The smoke detection module is configured to input the visible light image into a preset smoke detection model, so that the smoke detection model generates a segmentation mask of the smoke region in the visible light image based on the visible light image; wherein the smoke detection model is trained using a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke region in the visible light image sample;
[0092] The smoke annotation module is used to annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image;
[0093] The wildfire identification module is used to obtain a user's first question information; the first question information includes: whether a wildfire exists and a method for answering the question; the first question information and the smoke-annotated image are input into a preset multimodal model, so that the multimodal model generates a determination result on whether a wildfire exists in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image, in the method for answering the question in the first question information; wherein, the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether a wildfire exists.
[0094] In a preferred embodiment, the transmission corridor wildfire identification device further comprises: a smoke detection model training module;
[0095] The smoke detection model training module is used to obtain a number of first training samples; input each first training sample into the smoke detection model in sequence, and train the smoke detection model until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; calculates a first loss function value based on the segmentation mask of the predicted smoke area and the corresponding semantic label; and updates the smoke detection model based on the first loss function value.
[0096] In a preferred embodiment, the transmission corridor wildfire identification device further includes: a multimodal model training module;
[0097] The multimodal model training module is used to obtain a number of second training samples; input each second training sample into the multimodal model in sequence, and train the multimodal model until a preset second training number is reached; wherein, when the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; calculates the second loss function value based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label; and updates the multimodal model based on the second loss function value.
[0098] It should be noted that the embodiment of the device described above corresponds to the above-mentioned embodiment of the present invention, and it can implement the method for identifying wildfires in the power transmission corridor described in any one of the above-mentioned embodiments of the present invention. In addition, the embodiment of the above-mentioned device is merely schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the embodiment of the device provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement it without paying any creative work.
[0099] Based on the above method embodiment of the present invention, a corresponding electronic device embodiment is provided.
[0100] An embodiment of the present invention provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for identifying wildfires in a power transmission corridor described in any one of the present invention is implemented, or when the processor executes the computer program, the functions of the modules in the above-mentioned device embodiments are implemented.
[0101] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.
[0102] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.
[0103] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.
[0104] The memory can be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.
[0105] Based on the above method embodiment, the present invention provides a corresponding storage medium embodiment;
[0106] Another embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute any of the above-mentioned forest fire identification methods for the power transmission corridor of the present invention.
[0107] The storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.
[0108] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.
[0109] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for identifying wildfires in power transmission corridors, characterized in that: include: Obtain visible light images of the transmission corridor to be studied; Inputting the visible light image into a preset smoke detection model so that the smoke detection model generates a segmentation mask of the smoke region in the visible light image based on the visible light image; wherein the smoke detection model is trained using a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke region in the visible light image sample; Annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image; Obtaining a first question from a user; the first question includes: whether there is a wildfire and how to answer the question; The first question information and the smoke-annotated image are input into a preset multimodal model, so that the multimodal model generates a determination result on whether a wildfire exists in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image in the form of an answer to the question in the first question information; wherein the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether a wildfire exists.
2. The method for identifying wildfires in power transmission corridors according to claim 1, wherein: The smoke detection model is trained in the following way: Obtaining a number of first training samples; Each first training sample is input into the smoke detection model in sequence, and the smoke detection model is trained until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; based on the segmentation mask of the predicted smoke area and the corresponding semantic label, a first loss function value is calculated; and the smoke detection model is updated according to the first loss function value.
3. The method for identifying wildfires in power transmission corridors according to claim 2, wherein: The multimodal model is trained in the following way: Obtaining a number of second training samples; Each second training sample is input into the multimodal model in sequence, and the multimodal model is trained until a preset second training number is reached; wherein, each time the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label, the second loss function value is calculated; and the multimodal model is updated according to the second loss function value.
4. The method for identifying wildfires in power transmission corridors according to claim 3, wherein: The first training samples are sequentially inputted into the smoke detection model to train the smoke detection model until a preset first training number is reached, further comprising: For the visible light image samples in the first training samples of each batch, data augmentation processing is performed to generate updated visible light image samples; wherein the data augmentation processing includes any one or a combination of random flipping, random scaling, random translation, random rotation, random adjustment of image brightness, random addition of Gaussian noise and random cropping.
5. The method for identifying wildfires in power transmission corridors according to claim 4, wherein: After inputting the first question information and the smoke-annotated image into a preset multimodal model, so that the multimodal model generates a determination result of whether a wildfire exists in the transmission corridor to be studied in the form of an answer to the question in the first question information based on the user's first question information and the smoke-annotated image, the method further includes: Send an alarm message when there is a wildfire in the transmission corridor to be studied.
6. A wildfire identification device for a power transmission corridor, characterized in that: include: Image acquisition module, smoke detection module, smoke labeling module and wildfire identification module; The image acquisition module is used to acquire visible light images of the transmission corridor to be studied; The smoke detection module is configured to input the visible light image into a preset smoke detection model, so that the smoke detection model generates a segmentation mask of the smoke region in the visible light image based on the visible light image; wherein the smoke detection model is trained using a plurality of first training samples; each first training sample includes a visible light image sample of the transmission corridor and a semantic label corresponding to the visible light image sample; the semantic label is a segmentation mask of the smoke region in the visible light image sample; The smoke annotation module is used to annotate the visible light image according to the segmentation mask of the smoke area in the visible light image to generate a smoke annotated image; The wildfire identification module is used to obtain a user's first question information; the first question information includes: whether a wildfire exists and a method for answering the question; the first question information and the smoke-annotated image are input into a preset multimodal model, so that the multimodal model generates a determination result on whether a wildfire exists in the transmission corridor to be studied based on the user's first question information and the smoke-annotated image, in the method for answering the question in the first question information; wherein, the multimodal model is trained by a plurality of second training samples; each second training sample includes a smoke-annotated image sample, the user's second question information, and a wildfire identification label corresponding to the smoke-annotated image sample for determining whether a wildfire exists.
7. The wildfire identification device for a power transmission corridor according to claim 6, characterized in that: Also includes: Smoke Detection model training module; The smoke detection model training module is used to obtain a plurality of first training samples; Each first training sample is input into the smoke detection model in sequence, and the smoke detection model is trained until a preset first training number of times is reached; wherein, when the smoke detection model receives each first training sample, it outputs a segmentation mask of the predicted smoke area corresponding to the first training sample; based on the segmentation mask of the predicted smoke area and the corresponding semantic label, a first loss function value is calculated; and the smoke detection model is updated according to the first loss function value.
8. The wildfire identification device for a power transmission corridor according to claim 7, characterized in that: Also includes: Multimodal model training module; The multimodal model training module is used to obtain a plurality of second training samples; Each second training sample is input into the multimodal model in sequence, and the multimodal model is trained until a preset second training number is reached; wherein, each time the multimodal model receives a second training sample, it outputs the predicted wildfire identification situation corresponding to the second training sample; based on the predicted smoke and wildfire identification situation and the corresponding wildfire identification label, the second loss function value is calculated; and the multimodal model is updated according to the second loss function value.
9. An electronic device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for identifying wildfires in a power transmission corridor according to any one of claims 1 to 5 is implemented.
10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the method for identifying wildfires in a power transmission corridor as described in any one of claims 1 to 1 to 5.