Mountain fire prediction method and device for line corridor, equipment and storage medium

Abstract

The embodiment of the invention discloses a line corridor mountain fire prediction method and device, equipment and a storage medium. The method comprises the following steps: acquiring a fire image video library and a non-mountain fire background image video library; performing forest fire simulation according to the fire image video library and the non-forest fire background image video library to obtain a simulated forest fire smoke training set of the line corridor; training a preset neural network according to the simulated forest fire smoke training set and the non-forest fire background image video library of the line corridor to obtain a forest fire prediction model of the line corridor; and performing forest fire prediction based on the forest fire prediction model, and determining whether a fire occurs in the line corridor. Performing forest fire simulation on the basis of a non-forest fire background image video library by using a fire image video library to obtain a simulated forest fire smoke training set which contains rich forest fire data possibly occurring in line corridors; and the prediction result of the forest fire prediction model obtained by training the simulated forest fire smoke training set is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of wildfire prediction, and in particular to a wildfire prediction method and device, equipment and storage medium for a line corridor. Background Art

[0002] Wildfires spread quickly and cover a wide area, causing extremely high losses to power transmission line resources. Therefore, early detection of wildfires and reducing the losses caused by wildfires have become top priorities.

[0003] Wildfires usually occur in remote areas without supervision. Due to the long distance from the ground, cloud cover and lack of all-weather coverage in a given area, it usually takes several hours for the fire to be detected. By then, it is very likely that the transmission line has tripped or broken. Therefore, the current methods for preventing wildfires still have problems such as low efficiency and large errors in detection results. Summary of the invention

[0004] Based on this, it is necessary to propose a wildfire prediction method, device, equipment and storage medium for the line corridor to address the above problems, so as to improve the efficiency of wildfire prediction and effectively reduce the error of the prediction results.

[0005] To achieve the above objectives, the present application provides a first aspect of a wildfire prediction method for a line corridor, the method comprising:

[0006] Obtain a video library of fire images and a video library of non-wildfire background images;

[0007] Perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor;

[0008] Training a preset neural network according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model for the line corridor;

[0009] A wildfire prediction is performed based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

[0010] Furthermore, the acquisition of the fire image video library and the non-wildfire background image video library specifically includes:

[0011] Collect images and / or videos of real fires that occurred in history to generate a fire image and video library;

[0012] Background images and / or videos of the area where the line corridor is located at different time nodes when no wildfire occurs are collected to generate a non-wildfire background image video library of the line corridor.

[0013] Furthermore, performing a wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor specifically includes:

[0014] Extracting a plurality of mountain fire smoke feature types and reference values ​​of the various mountain fire smoke feature types based on the fire image video library;

[0015] A wildfire simulation is performed on the non-wildfire background image video library according to the wildfire smoke feature type and the reference value of the wildfire smoke feature type to obtain a simulated wildfire smoke training set for the line corridor.

[0016] Further, performing a mountain fire simulation on the non-mountain fire background image video library according to the mountain fire smoke feature type and the reference value of the mountain fire smoke feature type to obtain a simulated mountain fire smoke training set for the line corridor specifically includes:

[0017] Combining one or more wildfire smoke feature types, using reference values ​​of the wildfire smoke feature types as variables, to obtain a number of combined factor samples;

[0018] Generate several groups of three-dimensional simulated image video sets of wildfire smoke according to the combined factor samples;

[0019] The three-dimensional simulated image video set of the wildfire smoke is inserted into the non-wildfire background image video library for rendering to obtain a simulated wildfire smoke training set of the line corridor.

[0020] Furthermore, the preset neural network is trained according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain the wildfire prediction model of the line corridor, which specifically includes:

[0021] Performing a first round of training on the neural network according to the non-wildfire background image video set to obtain an initial wildfire prediction model;

[0022] The simulated wildfire smoke training set of the line corridor is added to the initial wildfire prediction model for a second round of training to obtain a target wildfire prediction model, wherein the input of the wildfire prediction model is an image video, and the output is the probability of a wildfire occurring in the input image video.

[0023] Further, the performing of wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor specifically includes:

[0024] Collecting actual image videos of the line corridor;

[0025] Performing frame division and / or image segmentation on the actual image video to obtain a plurality of image blocks;

[0026] Inputting all image blocks into the wildfire prediction model to perform wildfire prediction, and obtaining the probability of wildfire occurrence in each image block output by the wildfire prediction model;

[0027] Whether a fire occurs in the line corridor is determined according to the probability of a wildfire occurring in all image blocks.

[0028] Furthermore, after performing a wildfire prediction based on the wildfire prediction model and determining whether a fire occurs in the line corridor, the method further includes:

[0029] According to the probability of wildfire occurrence in each image block, determine the target image block where the wildfire phenomenon exists;

[0030] The wildfire is located according to the position of the target image block in the actual image video to determine the source of the wildfire.

[0031] To achieve the above-mentioned purpose, the second aspect of the present application provides a wildfire prediction device for a line corridor, the device comprising a data acquisition unit, a model training unit and a wildfire prediction unit;

[0032] The data acquisition unit is used to obtain a fire image video library and a non-wildfire background image video library;

[0033] The model training unit is used to perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor;

[0034] Training a preset neural network according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model for the line corridor;

[0035] The wildfire prediction unit is used to perform wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

[0036] To achieve the above-mentioned purpose, the third aspect of the present application provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the processor executes the steps of the method described in the first aspect.

[0037] To achieve the above-mentioned purpose, the fourth aspect of the present application provides a computer device, including a memory and a processor, characterized in that the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described in the first aspect.

[0038] The embodiments of the present invention have the following beneficial effects:

[0039] The embodiment of the present invention proposes a method for predicting wildfires in a line corridor, the method comprising: obtaining a fire image video library and a non-wildfire background image video library; performing wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor; training a preset neural network based on the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model for the line corridor; performing wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor. Wildfire simulation is performed using the fire image video library on the basis of the non-wildfire background image video library to obtain a simulated wildfire smoke training set, the simulated wildfire smoke training set contains abundant wildfire data that may occur in the line corridor, so that the prediction result of the wildfire prediction model trained using the simulated wildfire smoke training set is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] in:

[0042] Figure 1 A schematic flow chart of a method for predicting wildfires in a line corridor according to an embodiment of the present invention;

[0043] Figure 2 A three-dimensional simulated image generated in an embodiment of the present invention;

[0044] Figure 3 A simulated mountain fire smoke image in an embodiment of the present invention;

[0045] Figure 4 It is a structural block diagram of a wildfire prediction device for a line corridor in an embodiment of the present invention;

[0046] Figure 5 1 is a diagram showing the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Installing monitoring systems on power towers is a relatively effective method for early detection of wildfires. However, since wildfires usually occur in remote areas and there is almost no supporting infrastructure, the maintenance cost of monitoring systems on towers in remote areas is high, and there are blind spots in monitoring wildfires. In addition, the cameras in the monitoring system are stationary, which greatly limits the monitoring range.

[0049] To overcome this shortcoming, the embodiment of the present invention uses drones to inspect line corridors for wildfires. The drones have strong maneuverability and can cover a wide range of terrain in a short time. In the embodiment of the present invention, drones are used to patrol transmission line wildfires and collect on-site videos of line corridors. The acquired videos are processed and analyzed in real time in combination with deep neural networks to detect and warn of wildfires as early as possible, thereby reducing the losses caused by wildfires. By constructing a deep neural network that can be executed in a lightweight manner and can be deployed on conventional mobile hardware platforms, the problems of high hardware requirements for neural network operations and low efficiency of edge computing are solved.

[0050] An embodiment of the present invention proposes a wildfire prediction system for a line corridor based on a line patrol drone, which can be used to detect wildfires on a fixed or mobile platform. The fixed platform can be a line communication device, and the mobile platform can be a mobile communication terminal, such as a smart phone, a tablet computer, a portable computer, etc.

[0051] The physical components of the wildfire prediction system include: drones, cameras, processors, IoT communication devices, servers, computers, and any other devices with physical support and digital capture and / or processing data capabilities. Select the appropriate drone model based on the shooting requirements, considering factors such as endurance, signal transmission stability, flight speed, load capacity, etc. Ensure that the drone has a stable flight control system to cope with flight missions under complex environmental conditions. Select a high-resolution, high-sensitivity camera to ensure that the video footage is clear and has high color reproduction. Consider the camera's anti-shake function to reduce the impact of shaking during flight on video quality. Cameras with night vision, infrared and other functions can also be selected as needed to meet the shooting needs under different light conditions. The camera is deployed on the drone to capture video information from the monitoring area where the line corridor is located, and the video information is transmitted to the processor, server, etc. through the IoT communication device for information storage, information distribution and information processing.

[0052] The present invention also proposes a wildfire prediction method for line corridors, which can be deployed on a local or background server for monitoring wildfires, such as a video monitoring platform on a line tower or a line patrol drone, to identify flames and smoke columns in live videos in the monitoring area provided by hardware devices. Transfer learning, fine-tuning convolutional neural networks, and block-based detection are used to implement an additional adversarial neural network (GAN network) classifier that separates wildfire images from conventional image videos to predict wildfires in the monitoring area.

[0053] For details, please refer to Figure 1 , Figure 1 : is a flow chart of a method for predicting wildfires in a line corridor in an embodiment of the present invention, the method comprising:

[0054] Step 110: Acquire a fire image video library and a non-wildfire background image video library.

[0055] In the embodiment of the present invention, when a fire occurs, the combustible material will produce a large amount of gas and tiny particles during the combustion process. In the initial combustion stage of the fire, the combustible material begins to pyrolyze at high temperature, releasing volatile gases and solid carbon particles. As the fire spreads, these gases and carbon particles react with oxygen in the air to form a large amount of smoke. Therefore, the smoke characteristics are the key characteristics of a volcano. Based on this, the present invention chooses to analyze based on the characteristics of mountain fire smoke and detect mountain fires in the line corridor.

[0056] The fire image video library contains a large number of real fire image videos with obvious fire smoke characteristics that have occurred in history, which are used to train the wildfire smoke characteristics in the neural model. Smoke characteristics include at least: (1) Color characteristics: The color of smoke can reflect its composition and combustion state. For example, white smoke usually indicates a lower temperature, which may be a thick fog caused by the evaporation of water due to residual heat after a large amount of water is poured for rescue; black smoke usually indicates the highest temperature and the most intense fire; gray smoke may indicate smoldering or high temperature waiting for heating; yellow-green smoke is the most toxic chemical yellow-green poisonous fog. (2) Density and fluidity: The density of smoke is affected by temperature and composition. In the early stage of a fire, the density of smoke may be small, but as the fire spreads and the temperature rises, the density of smoke will gradually increase. In addition, smoke will continue to flow and spread in the building, making it difficult for the fire to spread and for people to evacuate. (3) Texture characteristics: Smoke is a turbulent phenomenon in the combustion process, and its texture characteristics are feather-like or cloud-like. These texture characteristics change continuously at different locations in space and repeat in the same area. By collecting a large number of real fire image videos, we can analyze and obtain richer smoke characteristics in order to improve the accuracy of wildfire prediction results.

[0057] The non-wildfire background image video library contains a large number of on-site image video libraries of line corridors when no fire occurs, which are used to contrast with the smoke characteristics when a fire occurs, so as to train the neural model so that the obtained classifier can better distinguish whether a wildfire occurs.

[0058] In one embodiment of the present invention, the image videos in the non-wildfire background image video library can be acquired based on drones and cameras. The drones patrol the area where the line corridor is located along a specified route, take photos and videos of the scene, and collect image videos when no fire occurs in the line corridor.

[0059] Step 120: Perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor.

[0060] In an embodiment of the present invention, the non-wildfire background image video library of the line corridor is used as the background image video, and the fire image video library and the non-wildfire background image video library are integrated to generate different types of simulated wildfire smoke image videos of the line corridor, thereby obtaining a simulated wildfire smoke training set of the line corridor. The simulated wildfire smoke training set is used to simulate the smoke characteristics when a wildfire occurs in the line corridor.

[0061] Step 130: Train a preset neural network based on the simulated wildfire smoke training set and the non-wildfire background image video library of the line corridor to obtain a wildfire prediction model for the line corridor.

[0062] The simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library collected in the early stage are used to form a training set to train the neural network in a targeted manner, and a wildfire prediction model for the line corridor is obtained.

[0063] Step 140: Perform wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

[0064] In the embodiment of the present invention, the line corridor is inspected regularly, and the on-site image video of the area where the line corridor is located is collected, and the on-site image video is input into the wildfire prediction model to obtain the mountain result output by the wildfire prediction model, and determine whether a wildfire occurs in the line corridor according to the output result. Real-time edge computing of wildfire monitoring is realized by drones, and a computing platform equipped with a low-power embedded processor as the core can monitor the fire situation in the line corridor and assist the operation and maintenance personnel to judge the wildfire situation in the line corridor in real time according to the specific situation.

[0065] The wildfire prediction method for the line corridor of the present invention is based on a simulated wildfire smoke training set and a non-wildfire background image video library of a specific line corridor, and adopts targeted training of a deep neural network to obtain a wildfire prediction model to improve the recognition efficiency and accuracy of wildfires.

[0066] In one embodiment of the present invention, obtaining a fire image video library and a non-wildfire background image video library specifically includes:

[0067] Step 111, collect images and / or videos of real fires that have occurred in history and generate a fire image and video library.

[0068] Step 112, collecting background images and / or videos of the area where the line corridor is located at different time nodes when no wildfire occurs, and generating a non-wildfire background image video library of the line corridor.

[0069] Since smoke is a turbulent phenomenon in the combustion process, its texture features are feather-like or cloud-like. These texture features keep changing in different positions in space and repeat in the same area. In other words, when a fire occurs, the smoke is always moving and changing. Therefore, the occurrence of the fire can be observed by obtaining multiple continuous image sequences and videos.

[0070] Based on this, the embodiment of the present invention selects to obtain image sequences and videos of real fires that occurred in history to generate a fire image video library. The non-wildfire background image video library is created by using drones to patrol the line corridors along the line to collect image sequences and videos under different circumstances at different time nodes. For example, considering the influence of ambient light intensity, multiple background image videos are obtained for the background images of the same area at different time periods and under different environmental conditions.

[0071] In one embodiment of the present invention, a training data set is created using a digitally generated simulated wildfire smoke image video, and a neural network deep learning is performed in combination with a non-wildfire background image video library to ensure that the neural network trained by the training set can accurately identify the wildfire status in the area where the line corridor is located. Specifically, a wildfire simulation is performed based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor, which specifically includes:

[0072] Step 121. Extract several wildfire smoke feature types and reference values ​​of various wildfire smoke feature types based on the fire image video library.

[0073] In the embodiment of the present invention, the training set focuses on the smoke in the early stage of a fire, so as to timely discover the fire situation in the line corridor and minimize the loss.

[0074] There are many factors that affect the shape of smoke. By conducting smoke analysis on historical fire cases, we can extract a variety of wildfire smoke characteristic types and reference values ​​for each wildfire smoke characteristic type, such as the range of variation.

[0075] In one embodiment of the present invention, the wildfire smoke feature types may include at least smoke color, smoke height, smoke width, smoke inclination, camera height and visibility, etc.

[0076] Step 122, perform wildfire simulation on the non-wildfire background image video library according to the wildfire smoke feature type and the reference value of the wildfire smoke feature type to obtain a simulated wildfire smoke training set for the line corridor.

[0077] Due to terrain and wind deviation, the smoke in the line corridor is always moving, and the smoke moves from one area of ​​the image / video frame to another in the video or image sequence. In view of this feature, a real smoke video image simulated by 3D animation is generated based on the wildfire smoke feature type and the reference value of the wildfire smoke feature type, and the real smoke video image is inserted into the collected background image to create a simulated wildfire smoke training set in the line corridor patrol area.

[0078] In one embodiment of the present invention, a non-mountain fire background image video library is subjected to a mountain fire simulation according to the mountain fire smoke feature type and the reference value of the mountain fire smoke feature type to obtain a simulated mountain fire smoke training set of the line corridor, specifically including:

[0079] Step 1221. Based on one or more wildfire smoke feature types, a combination is performed, and a reference value of the wildfire smoke feature type is used as a variable to obtain several combination factor samples.

[0080] Step 1222, generate several groups of three-dimensional simulated image video sets of wildfire smoke based on the combined factor samples.

[0081] Step 1223, insert the three-dimensional simulated image video set of wildfire smoke into the non-wildfire background image video library for rendering, and obtain the simulated wildfire smoke training set of the line corridor.

[0082] In an embodiment of the present invention, by taking six categories of smoke color, smoke height, smoke width, smoke inclination, camera height, and visibility as variables and changing the reference value size of each type of variable, multiple groups of combined factor samples are obtained, and different simulated wildfire smoke image videos are generated based on the combined factor samples.

[0083] For details, please refer to Table 1, which contains five different combination factor samples A, B, C, D, and E. The reference values ​​of wildfire smoke characteristics in each group of samples are different.

[0084] Table 1 Sample examples of different combination factors

[0085]

[0086]

[0087] After obtaining the combined factor samples, a three-dimensional simulated image video is generated according to each combined factor sample, which is used for simulated image videos of wildfire smoke characteristics under different wildfire smoke characteristics. Figure 2 , Figure 2 It is a three-dimensional simulated image generated in an embodiment of the present invention.

[0088] Insert the 3D simulated image video set into the non-wildfire background image video library for rendering to obtain the simulated wildfire smoke training set of the line corridor. Figure 3 , Figure 3 This is a simulated mountain fire smoke image in an embodiment of the present invention.

[0089] The simulated wildfire smoke training set generated by rendering combines each non-wildfire background image video with each 3D simulated image video, scales the 3D simulated image video from the camera perspective to simulate the changes in wildfire smoke as the drone flies, starts fusion rendering in the first frame of the non-wildfire background image video, and performs fusion rendering in subsequent frames following the movement of the camera.

[0090] In one embodiment of the present invention, a preset neural network is trained according to a simulated wildfire smoke training set of a line corridor to obtain a wildfire prediction model of the line corridor, specifically including:

[0091] Step 131, perform the first round of training on the neural network based on the non-wildfire background image video set to obtain an initial wildfire prediction model.

[0092] Step 132, add the simulated wildfire smoke training set of the line corridor to the initial wildfire prediction model for a second round of training to obtain the target wildfire prediction model. The input of the wildfire prediction model is the image video, and the output is the probability of wildfire occurring in the input image video.

[0093] Since wildfire smoke has no specific shape or significant features, the basic idea of ​​smoke recognition is to regard smoke as an abnormal event or specific object in the observation scene. The present invention uses a deep neural network to form a classifier to identify whether the image video contains abnormal events or specific objects, so as to distinguish between normal image videos without wildfire smoke and wildfire image videos containing wildfire smoke.

[0094] There are two training stages in the training process of the neural network. In the first stage, the neural network is first trained using the noise vector added to the image from the non-wildfire background image video set. Since smoke and flame images are not in the non-wildfire background image video set, the probability value of the normal video scene of the line corridor output by the neural network is lower than 0.5, and the probability value of the video containing smoke and flames is higher than 0.5. During the training process of the network, the function of the output probability value is updated according to the change of the random gradient, as shown in the following formula:

[0095]

[0096] In the formula, x i and z i are the i-th non-wildfire background image video data and noise vector, G is the input noise vector z i Generator for generating “fake” images. The vector θd contains the parameters in the neural network, and M is the total number of non-wildfire background image video data.

[0097] In order to improve the recognition accuracy, the initial wildfire prediction model is trained in the second stage by using stochastic gradient to fine-tune the neural network, and the output probability value is calculated as shown in the following formula:

[0098]

[0099] In the formula, y i represents the i-th image video in the simulated wildfire smoke training set, and L is the total number of image videos in the simulated wildfire smoke training set. The total number of image videos in the simulated wildfire smoke training set, L, is much larger than the total number of non-wildfire background image video data, M. During the training process, the parameters of the generator network in the neural network are updated in real time.

[0100] The embodiment of the present invention generates an adversarial network during the training process to distinguish between wildfire smoke and non-wildfire smoke image videos. At the same time, the configuration requirements can be reduced to improve the computational efficiency of the model so that it can be deployed in an embedded processor carried by a drone. In the present invention, an additive network operator can be used to replace the multiplication operator in the neural network to reduce the calculation process and improve the processing efficiency, because compared with addition in most processors, multiplication consumes much more energy.

[0101] After obtaining the target wildfire prediction model after two rounds of training, the actual image video of the line corridor collected in real time is input into the target wildfire prediction model. If the output probability of the target wildfire prediction model is higher than 0.5, it means that there is a fire in the line corridor, and an alarm signal needs to be issued in time, and the images and videos are stored for subsequent model training and updating.

[0102] Since smoke may exist in any part of an image or video frame, even at the edge, in order to ensure that smoke can be detected even if it appears at the edge of an image or video frame, in one embodiment of the present invention, a wildfire prediction is performed based on a wildfire prediction model to determine whether a fire occurs in a line corridor, specifically including:

[0103] Step 141. Collect actual image videos of the line corridor.

[0104] Specifically, the drone conducts apex cruises on the line corridor and collects actual image videos of the line corridor.

[0105] Step 142: Frame and / or image segmentation is performed on the actual image video to obtain a plurality of image blocks.

[0106] Specifically, the video is divided into frames to obtain several video frames, and then the video frame or image is segmented to obtain several image blocks. For example, the image or video frame is divided into 20 blocks in a 4×5 grid, in which the grid is configured in an overlapping manner, so if smoke occurs at the edge of a block, it will also exist in its adjacent block. The overlap between blocks can be a horizontal overlap, such as a portion of the right half of a block overlaps with a portion of the left half of its right adjacent block; a vertical overlap, such as a portion of the upper half of a block overlaps with a portion of the lower half of its upper adjacent block; a diagonal overlap, such as a portion of the upper right quarter of a block overlaps with a portion of the lower left quarter of its upper right adjacent block. The continuity between blocks is ensured by overlapping segmentation.

[0107] Step 143, input all image blocks into the wildfire prediction model for wildfire prediction, and obtain the probability of wildfire occurrence in each image block output by the wildfire prediction model.

[0108] Step 144, determine whether a fire occurs in the line corridor according to the probability of wildfire occurrence in all image blocks.

[0109] In one embodiment of the present invention, the number of image blocks where wildfires occur is counted, and if the proportion of the number in all image blocks exceeds a preset ratio threshold, it is considered that a fire has occurred in the line corridor.

[0110] In another embodiment of the present invention, if image blocks indicating a wildfire situation occur are concentrated together and their area is greater than a preset area threshold, it is considered that a fire occurs in the line corridor.

[0111] In a feasible embodiment of the present invention, after performing a wildfire prediction based on a wildfire prediction model to determine whether a fire has occurred in a line corridor, the method further includes: determining a target image block where a wildfire phenomenon exists based on the probability of a wildfire occurring in each image block; locating the wildfire based on the position of the target image block in the actual image video to determine the source of the wildfire.

[0112] Specifically, after confirming that a fire has occurred in a line corridor, the specific position of the image block where the wildfire occurred is determined in the complete image or video frame, and the fault is located in the line corridor according to the specific position to determine the source of the wildfire.

[0113] The embodiment of the present invention also provides a wildfire prediction device for a line corridor, which can be referred to Figure 4 , Figure 4 40 is a structural block diagram of a wildfire prediction device for a line corridor in an embodiment of the present invention. The device includes a data acquisition unit 401, a model training unit 402 and a wildfire prediction unit 403.

[0114] The data acquisition unit 401 is used to obtain a fire image video library and a non-wildfire background image video library.

[0115] The model training unit 402 is used to perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor.

[0116] The preset neural network is trained according to the simulated wildfire smoke training set of the line corridor to obtain the wildfire prediction model of the line corridor.

[0117] The wildfire prediction unit 403 is used to perform wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

[0118] The wildfire prediction device for the line corridor proposed in the present invention uses a fire image video library to perform wildfire simulation on the basis of a non-wildfire background image video library to obtain a simulated wildfire smoke training set. The simulated wildfire smoke training set contains rich wildfire data that may occur in the line corridor, so that the prediction results of the wildfire prediction model trained using the simulated wildfire smoke training set are more accurate.

[0119] Figure 5 FIG. 1 shows an internal structure diagram of a computer device in one embodiment of the present invention. The computer device may be a terminal or a system. Figure 5 As shown, the computer device includes a processor, a memory and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor may implement each step in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor may implement each step in the above method embodiment. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0120] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes each step in the above method embodiment.

[0121] In one embodiment, a computer-readable storage medium is proposed, which stores a computer program. When the computer program is executed by a processor, the processor executes each step in the above method embodiment. A person of ordinary skill in the art can understand that all or part of the processes in the above embodiment method can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium, and the program can include the processes of the embodiments of the above methods when executed. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0122] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting wildfires in a line corridor, characterized in that: The method comprises: Obtain a video library of fire images and a video library of non-wildfire background images; Perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor; Training a preset neural network according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model for the line corridor; A wildfire prediction is performed based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

2. The method according to claim 1, characterized in that The acquisition of the fire image video library and the non-wildfire background image video library specifically includes: Collect images and / or videos of real fires that occurred in history to generate a fire image and video library; Background images and / or videos of the area where the line corridor is located at different time nodes when no wildfire occurs are collected to generate a non-wildfire background image video library of the line corridor.

3. The method according to claim 2, characterized in that The method of performing a wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor specifically includes: Extracting a plurality of mountain fire smoke feature types and reference values ​​of the various mountain fire smoke feature types based on the fire image video library; A wildfire simulation is performed on the non-wildfire background image video library according to the wildfire smoke feature type and the reference value of the wildfire smoke feature type to obtain a simulated wildfire smoke training set for the line corridor.

4. The method according to claim 3, characterized in that The method of performing a mountain fire simulation on the non-mountain fire background image video library according to the mountain fire smoke feature type and the reference value of the mountain fire smoke feature type to obtain a simulated mountain fire smoke training set for the line corridor specifically includes: Combining one or more wildfire smoke feature types, using reference values ​​of the wildfire smoke feature types as variables, to obtain a number of combined factor samples; Generate several groups of three-dimensional simulated image video sets of wildfire smoke according to the combined factor samples; The three-dimensional simulated image video set of the wildfire smoke is inserted into the non-wildfire background image video library for rendering to obtain a simulated wildfire smoke training set of the line corridor.

5. The method according to claim 1, characterized in that The method of training a preset neural network according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model of the line corridor specifically includes: Performing a first round of training on the neural network according to the non-wildfire background image video set to obtain an initial wildfire prediction model; The simulated wildfire smoke training set of the line corridor is added to the initial wildfire prediction model for a second round of training to obtain a target wildfire prediction model, wherein the input of the wildfire prediction model is an image video, and the output is the probability of a wildfire occurring in the input image video.

6. The method according to claim 1, characterized in that The performing of wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor specifically includes: Collecting actual image videos of the line corridor; Performing frame division and / or image segmentation on the actual image video to obtain a plurality of image blocks; Inputting all image blocks into the wildfire prediction model to perform wildfire prediction, and obtaining the probability of wildfire occurrence in each image block output by the wildfire prediction model; Whether a fire occurs in the line corridor is determined according to the probability of a wildfire occurring in all image blocks.

7. The method according to claim 6, characterized in that After performing a wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor, the method further includes: According to the probability of wildfire occurrence in each image block, determine the target image block where the wildfire phenomenon exists; The wildfire is located according to the position of the target image block in the actual image video to determine the source of the wildfire.

8. A wildfire prediction device for a line corridor, characterized in that: The device includes a data acquisition unit, a model training unit and a wildfire prediction unit; The data acquisition unit is used to obtain a fire image video library and a non-wildfire background image video library; The model training unit is used to perform wildfire simulation based on the fire image video library and the non-wildfire background image video library to obtain a simulated wildfire smoke training set for the line corridor; Training a preset neural network according to the simulated wildfire smoke training set of the line corridor and the non-wildfire background image video library to obtain a wildfire prediction model for the line corridor; The wildfire prediction unit is used to perform wildfire prediction based on the wildfire prediction model to determine whether a fire occurs in the line corridor.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

Images