Real-time prediction method of fire scenes based on deep neural network

Abstract

The present invention relates to the field of fire prediction, and specifically discloses a real-time prediction method for fire scenes based on a deep neural network, including detecting accumulation features, smoke release features and smoldering features in a fire scene; determining that an area is a smoldering risk area in response to the simultaneous existence of accumulation features, smoke release features and smoldering features in the same area; verifying the smoldering risk area based on the appearance time of each feature; for the verified correct smoldering risk area, inputting the degree parameter of each feature into a neural network model, and outputting the deflagration risk level of the smoldering risk area; correcting the evacuation path based on the location and deflagration risk level of the smoldering risk area; detecting and analyzing areas where smoldering may turn into deflagration through a neural network model, and giving a prediction of the deflagration risk level, thereby providing data support and decision-making assistance for fire emergency response.

Description

Technical Field

[0001] The present invention relates to the field of fire prediction, and in particular to a real-time fire scene prediction method based on a deep neural network. Background Art

[0002] The prediction of fire development trends is of great significance for guiding the evacuation of on-site personnel and formulating fire-fighting strategies. In recent years, with the development of computer image processing, neural networks and deep learning related research, there have been relatively mature applications in the identification of fire hazards and the judgment of initial fires.

[0003] For example, the prior art discloses a method for planning the optimal escape route for fires based on a multimodal fusion algorithm, a target detection algorithm based on deep learning, and a multimodal trajectory prediction algorithm, which can accurately predict the trend of a fire, reduce the number of people who blindly choose escape routes due to the continuous expansion of the fire and the unknown trajectory of the fire, and have stronger predictability. The optimal escape route and the lowest temperature area are obtained through multiple point cameras on site.

[0004] However, the above scheme lacks prediction for possible deflagration in fire, which may cause risks in the planned escape routes due to deflagration and lacks sufficient reliability and safety. Summary of the invention

[0005] The purpose of the present invention is to provide a real-time prediction method for fire scenes based on deep neural networks to solve the problem that the prior art lacks prediction of possible deflagration phenomena in fires, resulting in the planned escape routes being at risk due to deflagration and lacking sufficient reliability and safety.

[0006] To this end, the present invention provides a real-time prediction method for fire scenes based on a deep neural network, the real-time prediction method for fire scenes based on a deep neural network includes detecting accumulation features, smoke release features, and smoldering features in a fire scene;

[0007] In response to the accumulation feature, the smoke release feature and the smoldering feature existing in the same area at the same time, determining that the area is a smoldering risk area;

[0008] Verify the smoldering risk area based on the occurrence time of each feature;

[0009] For the verified correct smoldering risk area, the degree parameters of each feature are input into the neural network model, and the deflagration risk level of the smoldering risk area is output;

[0010] Correction of evacuation routes based on the location of smoldering risk areas and the level of deflagration risk;

[0011] Among them, the smoke release characteristic satisfies that the smoke release amount in the area in the fire scene has an increasing trend, the smoldering characteristic satisfies that there are combustible materials inside the accumulation, and the deflagration risk level is used to characterize the triggering probability and triggering speed of deflagration.

[0012] As a preferred technical solution of the real-time prediction method of fire scenes based on deep neural networks, the determination of the accumulation characteristics includes:

[0013] Analyze the distribution density and height of objects in the area based on the images in the area;

[0014] If the distribution density of objects in the area exceeds a preset density threshold, and the height of the deposits shows a sudden increase or a continuous upward trend, it is determined that the area has the deposit feature.

[0015] As a preferred technical solution of the real-time prediction method for fire scenes based on deep neural networks, the determination of the accumulation characteristics also includes:

[0016] Compare the pre-stored material information in the scene with the deposits in the area;

[0017] The situation where the deposit is non-combustible is not determined as the deposit characteristic.

[0018] As a preferred technical solution of the real-time prediction method of fire scenes based on deep neural networks, the determination of the smoldering characteristics includes:

[0019] The image of the area is collected through an infrared binocular camera;

[0020] If there is an object covering the burning in the image, it is determined that the smoldering feature exists;

[0021] The temperature distribution of the deposit is obtained. If the temperature rises locally and the change gradient exceeds the gradient threshold, or the local temperature is higher than the temperature threshold, it is determined that the smoldering feature exists.

[0022] As a preferred technical solution of the real-time prediction method of fire scenes based on deep neural networks, the determination of the smoke release characteristics includes:

[0023] Determine the smoke coverage area in the image by an image processing algorithm;

[0024] If the smoke coverage area in the images within the historical period has shown an increasing trend, it is determined that the area has the smoke release characteristics.

[0025] As a preferred technical solution of the real-time prediction method of fire scenes based on deep neural networks, the verification of the smoldering risk area based on the appearance time of each feature includes:

[0026] Compare the appearance time of the smoke release characteristics and the accumulation characteristics. If the smoke release characteristics appear before the accumulation characteristics, it is determined that the analysis of the ignition risk area is incorrect.

[0027] As a preferred technical solution of the real-time prediction method of fire scenes based on deep neural networks, after verifying the smoldering risk area based on the appearance time of each feature, it also includes verifying the location of the smoldering risk area:

[0028] Determine the distance between each feature trigger position;

[0029] If the distance between any two features is greater than the distance threshold, it is determined that the analysis of the smoldering risk area is wrong.

[0030] As a preferred technical solution for the real-time prediction method of fire scenes based on deep neural networks, the degree parameters of each feature include the temperature and change gradient of the deposit, the smoke coverage area, the smoke rising speed and the accumulation range of the deposit.

[0031] As a preferred technical solution of the real-time prediction method for fire scenes based on deep neural networks, the real-time prediction method for fire scenes based on deep neural networks also includes establishing a neural network model for predicting deflagration risk:

[0032] Obtain training data from fire events, including the temperature of the deposit, the temperature gradient, the smoke coverage area, the smoke rising speed, the deposit accumulation range, and whether the deposit deflagrates and the values ​​of the training data corresponding to the deflagration time;

[0033] Inputting the training data into the input layer of the neural network model;

[0034] Accelerate model convergence and handle nonlinear features through ReLU;

[0035] The sigmoid function is used to compress the output to the range of 0 to 1, which comprehensively represents the probability and triggering speed of the deflagration risk.

[0036] As the preferred technical solution for the real-time prediction method of fire scenes based on deep neural networks, the deflagration risk level for determining the smoldering risk includes:

[0037] Input the degree parameters of each characteristic of the smoldering risk area into the neural network model to obtain the output results;

[0038] The deflagration risk level is divided according to the output results.

[0039] The beneficial effects of the present invention are:

[0040] The present invention provides a real-time prediction method for fire scenes based on a deep neural network. The real-time prediction method for fire scenes based on a deep neural network can more comprehensively identify potential risks in fire scenes by introducing a combination of multiple features, including accumulation features, smoke release features, and smoldering features. Traditional fire prediction methods may ignore smoldering features and the risk of deflagration caused by them, while the present invention can analyze these features through a neural network model and give an accurate prediction of the deflagration risk level, thereby providing more reliable information for fire emergency response.

[0041] In particular, the present invention ensures more accurate identification of smoldering risk areas and reduces the probability of misjudgment by verifying each feature in the fire scene in time and space (such as the time sequence of accumulation features and smoke release features, and the distance between feature triggering positions). This multi-level analysis based on time and space verification can adjust the prediction results in real time and improve the accuracy of fire prediction.

[0042] In particular, the deflagration phenomenon in fire scenes is highly sudden, and there is a lack of effective prediction and prevention measures in the prior art. The present invention uses a neural network model to classify the deflagration risk level in real time, and dynamically adjusts the evacuation path and emergency strategy according to the severity of the characteristics, thereby improving the adaptability and safety of emergency response in fire scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a method for real-time prediction of fire scenes based on a deep neural network in an embodiment of the present invention;

[0044] Figure 2 A flow chart for determining accumulation characteristics in an embodiment of the present invention;

[0045] Figure 3 A flow chart for determining smoldering characteristics in an embodiment of the present invention;

[0046] Figure 4 The flowchart of determining smoke release characteristics in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The technical solution of the present invention will be described clearly and completely below 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. 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] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0050] See also Figure 1 As shown, this embodiment provides a real-time prediction method for fire scenes based on a deep neural network, and the real-time prediction method for fire scenes based on a deep neural network includes:

[0051] Step S1, detecting accumulation features, smoke release features and smoldering features in a fire scene;

[0052] Step S2, in response to the presence of accumulation characteristics, smoke release characteristics and smoldering characteristics in the same area, determining that the area is a smoldering risk area;

[0053] Step S3, verifying the smoldering risk area based on the appearance time of each feature;

[0054] Step S4, for the verified correct smoldering risk area, input the degree parameter of each feature into the neural network model, and output the deflagration risk level of the smoldering risk area;

[0055] Step S5, correcting the evacuation route based on the location of the smoldering risk area and the deflagration risk level;

[0056] Among them, the smoke release characteristic satisfies that the smoke release amount in the area in the fire scene has shown an increasing trend, the smoldering characteristic satisfies that there are combustible materials inside the accumulation, and the deflagration risk level is used to characterize the triggering probability and triggering speed of deflagration.

[0057] In the above embodiment, by introducing a combination of multiple features, including accumulation features, smoke release features and smoldering features, the potential risks in the fire scene can be more comprehensively identified. Traditional fire prediction methods may ignore the smoldering features and the risk of deflagration caused by them, while this embodiment can analyze these features through a neural network model and give an accurate prediction of the deflagration risk level, thereby providing more reliable information for fire emergency response. By verifying each feature in the fire scene in time and space (such as the time sequence of accumulation features and smoke release features, and the distance between feature triggering positions), it is ensured that the identification of the smoldering risk area is more accurate and the probability of misjudgment is reduced. This multi-level analysis based on time and space verification can adjust the prediction results in real time and improve the accuracy of fire prediction. In addition, the deflagration phenomenon in the fire scene has a strong suddenness, and there is a lack of effective prediction and prevention measures for it in the prior art. The present invention divides the deflagration risk level in real time through a neural network model, and dynamically adjusts the evacuation path and emergency strategy according to the severity of the features, thereby improving the adaptability and safety of emergency response in the fire scene.

[0058] In detail, the determination of the smoldering risk zone in this embodiment has a good risk avoidance effect. Smoldering refers to a combustion state in which the combustion process is relatively slow, the combustion speed is low, and the temperature is relatively low under relatively hypoxic conditions. During the smoldering process, the combustion is mainly maintained by the heat generated by the decomposition reaction of the substance and the low-temperature oxidation. Since the smoldering temperature is low, the heat cannot be fully diffused to the surroundings, which may cause the substance to gradually accumulate. As the heat accumulates and reaches a certain threshold, the smoldering may suddenly turn into a deflagration.

[0059] Deflagration refers to the phenomenon that burning materials release energy quickly, accompanied by violent reactions and flame propagation. Under smoldering conditions, when the temperature gradually rises to a certain critical point, the combustibles in the smoldering area reach the critical temperature and react rapidly, and after mixing with the air, they suddenly enter the violent combustion stage, forming a deflagration;

[0060] Specifically, see Figure 2 As shown, the determination of the accumulation characteristics includes:

[0061] Step S101, analyzing the distribution density and distribution height of objects in the area according to the images in the area;

[0062] Step S102: If the distribution density of objects in the area exceeds a preset density threshold, and the height of the deposits shows a sudden increase or a continuous upward trend, it is determined that the area has an accumulation feature.

[0063] The determination of accumulation characteristics also includes:

[0064] Compare the pre-stored material information in the scene with the deposits in the area;

[0065] If the deposit is non-combustible, it is not considered as a deposit characteristic.

[0066] Specifically, see Figure 3 As shown, the determination of smoldering characteristics includes:

[0067] Step S111, collecting images of the area through an infrared binocular camera;

[0068] Step S112, if there is an object covering the burning in the image, it is determined that there is a smoldering feature;

[0069] Step S113, obtain the temperature distribution of the deposit, if the temperature rises locally and the change gradient exceeds the gradient threshold, or the local temperature is higher than the temperature threshold, it is determined that there is a smoldering feature. In implementation, the gradient threshold, temperature threshold and preset density threshold need to be determined in combination with the scene (such as area, fire progress, whether it is closed), and those skilled in the art can set them according to empirical data, which will not be repeated here.

[0070] See also Figure 4 As shown, the determination of smoke release characteristics includes:

[0071] Step S121, determining the smoke coverage area in the image by using an image processing algorithm;

[0072] Step S122: If the smoke coverage area in the image within the historical period (i.e., from the start of the fire to the current moment) has shown an increasing trend, it is determined that the area has smoke release characteristics.

[0073] In detail, the scenario of smoldering to deflagration can be covered by the three features of this embodiment:

[0074] The accumulation feature refers to the accumulation state of combustible materials in a fire scene. If combustible materials are accumulated in a large area with a high distribution density, it means that the risk of smoldering in the area is relatively high. The heat in the accumulation is difficult to dissipate, and the temperature may gradually increase, promoting the occurrence of smoldering. Therefore, by detecting the range, distribution density, and height changes of the accumulation, the possibility of smoldering can be predicted. For example, if the distribution density of objects in the area exceeds the threshold and the accumulation height is on an upward trend, it means that a smoldering hotspot may be formed, especially in a closed or semi-closed environment.

[0075] The smoke release characteristics can reflect the accumulation of incomplete combustion products during the combustion process. During the smoldering process, due to incomplete combustion, more smoke will be released, and the smoke release amount usually increases gradually. If the smoke release in the same area exceeds the threshold and continues to increase, it indicates that the burning materials in the area are decomposing in large quantities and generating combustible gases, which promotes the transition from smoldering to deflagration.

[0076] The characteristics of smoldering include the temperature changes in the local area and the combustion state inside the deposit. There are two identifiable situations, one is the situation where the burning material is covered, and the other is the situation where heat release is detected. By monitoring the temperature changes in the area with an infrared camera, especially the abnormal increase in local temperature or the sharp increase in the temperature change gradient, potential smoldering areas can be discovered in time. If the temperature inside the deposit rises under the premise of the existence of the smoke release characteristics, it means that there is a trend of turning into deflagration, and if the gas release volume increases too much and is now on a decreasing trend, but the temperature is rising at this time, it further indicates that the process of smoldering turning into deflagration is in progress and needs to be avoided in time.

[0077] The reaction time is the most important in the fire process, so the smoldering risk area needs to be verified to avoid erroneous analysis and delay the escape and disaster relief process. Therefore, the smoldering risk area is verified based on the appearance time of each feature, including:

[0078] Compare the appearance time of smoke release characteristics and accumulation characteristics. If the smoke release characteristics appear before the accumulation characteristics, it is determined that the analysis of the ignition risk area is wrong. It can be understood that the appearance of smoke release characteristics before accumulation indicates that the smoke release is not caused by accumulation, so there is no possibility of smoldering turning into deflagration.

[0079] After verifying the smoldering risk area based on the occurrence time of each feature, the location of the smoldering risk area is also verified:

[0080] Determine the distance between each feature trigger position;

[0081] If the distance between any two features is greater than the distance threshold, it is determined that the analysis of the smoldering risk area is wrong. It can be understood that if the smoke release feature is far away from the accumulation, it means that the smoke is not released by the accumulation and the smoldering risk does not exist. In practice, the distance threshold can be determined based on the fire coverage area.

[0082] Specifically, the degree parameters of each feature include the temperature and gradient of the deposit, the area covered by smoke, the speed of smoke rise, and the accumulation range of the deposit. The temperature of the deposit is an important indicator for judging the risk of smoldering or deflagration in a fire scene. One of the characteristics of smoldering is that the temperature in a local area rises significantly, while deflagration is often accompanied by a sharp change in the temperature of the deposit. Therefore, the temperature of the deposit can directly reflect the combustion process and the potential danger of the fire. The temperature change gradient can help determine whether the temperature distribution is uniform. In the process of smoldering to deflagration, the temperature in the local area usually rises sharply. Excessive temperature gradient indicates that the fire may spread faster or enter the deflagration stage. Therefore, by monitoring the gradient of temperature change, the severity and development trend of the fire can be identified in advance. Smoke release is one of the direct signs of fire development, especially in the smoldering stage, incomplete combustion will produce a lot of smoke. The increase in smoke coverage usually means the spread of fire. By detecting the range of smoke coverage, the size of the space affected by the fire can be inferred and the degree of danger of different areas can be evaluated. At the same time, the accumulation of smoke may bring the risk of deflagration, so the smoke coverage area is also an important parameter for judging the fire risk level. The rising speed of smoke reflects the intensity of the fire. Usually in the early stage of a fire, the rising speed of smoke is relatively slow. As the combustion intensifies, the rising speed of smoke will significantly accelerate. Especially in the stage of smoldering to deflagration, smoke will rise rapidly due to the intense combustion. Therefore, by monitoring the rising speed of smoke, the development momentum of the fire can be evaluated, the possibility of further expansion of the fire can be predicted, and emergency evacuation planning can be carried out. The scope of the accumulation directly affects the quality of the combustion materials of the fire and the ability of the fire to spread. Larger accumulations release more heat when burning, and the fire will spread faster. The scope of the accumulation can also reflect the potential smoldering area. An overly large accumulation area may form multiple fire points, exacerbating the development of the fire. Therefore, the size of the accumulation is an important parameter for judging the risk of fire spread.

[0083] As a preferred technical solution of the real-time prediction method for fire scenes based on deep neural networks, the real-time prediction method for fire scenes based on deep neural networks also includes establishing a neural network model for predicting deflagration risk:

[0084] Obtain training data from fire events, including the temperature of the deposit, the temperature gradient, the smoke coverage area, the smoke rising speed, the deposit accumulation range, and whether the deposit deflagrates and the values ​​of the training data corresponding to the deflagration time;

[0085] Input the training data into the input layer of the neural network model;

[0086] Accelerate model convergence and handle nonlinear features through ReLU;

[0087] The sigmoid function is used to compress the output to the range of 0 to 1, which comprehensively characterizes the probability and triggering speed of the risk of deflagration. In detail, in the hidden layer of the neural network, the ReLU (Rectified Linear Unit) activation function is used to process the nonlinear features of the input. The main function of the ReLU function is to speed up the convergence of the model while processing the nonlinear features in the data. ReLU directly maps negative values ​​to zero and keeps positive values ​​as they are, which helps prevent the gradient vanishing problem, especially when processing large-scale data sets, it can accelerate the model training process and improve the stability of the model. The output layer of the model uses the sigmoid activation function to compress the output value to the range of 0 to 1. This is because it is necessary to comprehensively characterize the probability and triggering speed of the risk of fire deflagration. The output value of the sigmoid function can be interpreted as the probability value of deflagration, while taking into account the change in triggering speed. By limiting the output of the model to this interval, it is convenient to divide the deflagration level during fire prediction, and ultimately provide a reliable basis for emergency decision-making for fires. Through the training of the neural network, the model can learn the complex relationship between various input features. The output value in the range of 0 to 1 can be directly used to characterize the risk of deflagration. If the output value is close to 1, it means that the possibility of deflagration in the area is high and may break out soon; if the output value is close to 0, it means that the risk of deflagration is low. The output of the model can serve as the core basis for real-time prediction of fire scenes and provide guidance for further emergency response.

[0088] As the preferred technical solution for the real-time prediction method of fire scenes based on deep neural networks, the deflagration risk level for determining the smoldering risk includes:

[0089] Input the degree parameters of each characteristic of the smoldering risk area into the neural network model to obtain the output results;

[0090] The deflagration risk level is divided according to the output results. The risk level division can be combined with the actual scenario. In actual use, different avoidance measures can be taken for the smoldering risk area based on the level division (such as reminding to stay away, reminding to pass quickly, modifying the route, etc.). For example, the output value can be divided into three levels:

[0091] Low risk: 0≤output value<0.33;

[0092] Medium risk: 0.33≤output value<0.66;

[0093] High risk: 0.66≤output value≤1.

[0094] The block diagram in the accompanying drawings illustrates the possible architecture, function and operation of the device, method and computer program product according to various embodiments of the present application. In this regard, each box in the block diagram can represent a part of a module, a program segment or a code, and the part of the module, the program segment or the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram, and the combination of the boxes in the block diagram and / or the flow chart can be implemented with a dedicated hardware-based device that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0095] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A real-time fire scene prediction method based on deep neural network, characterized in that: include: Detect accumulation features, smoke release features, and smoldering features in fire scenes; In response to the accumulation feature, the smoke release feature and the smoldering feature existing in the same area at the same time, determining that the area is a smoldering risk area; Verify the smoldering risk area based on the occurrence time of each feature; For the verified correct smoldering risk area, the degree parameters of each feature are input into the neural network model, and the deflagration risk level of the smoldering risk area is output; Correction of evacuation routes based on the location of smoldering risk areas and the level of deflagration risk; The smoke release characteristic satisfies that the smoke release amount in the area of ​​the fire scene has an increasing trend, the smoldering characteristic satisfies that there are combustibles inside the accumulation, and the deflagration risk level is used to characterize the triggering probability and triggering speed of deflagration; The determination of the accumulation characteristics includes: Analyze the distribution density and height of objects in the area based on the images in the area; If the distribution density of objects in the area exceeds a preset density threshold, and the height of the deposits shows a sudden increase or a continuous upward trend, it is determined that the area has the deposit feature; The determination of the accumulation characteristics also includes: Compare the pre-stored material information in the scene with the deposits in the area; The situation where the deposit is non-combustible is not determined as the deposit characteristic.

2. The method for real-time fire scene prediction based on deep neural network according to claim 1 is characterized in that: The determination of the smoldering characteristics includes: The image of the area is collected through an infrared binocular camera; If there is an object covering the burning in the image, it is determined that the smoldering feature exists; The temperature distribution of the deposit is obtained. If the temperature rises locally and the change gradient exceeds the gradient threshold, or the local temperature is higher than the temperature threshold, it is determined that the smoldering feature exists.

3. The method for real-time fire scene prediction based on deep neural network according to claim 2 is characterized in that: The determination of the smoke release characteristics includes: Determine the smoke coverage area in the image by an image processing algorithm; If the smoke coverage area in the images within the historical period has shown an increasing trend, it is determined that the area has the smoke release characteristics.

4. The method for real-time fire scene prediction based on deep neural network according to claim 3 is characterized in that: The verification of the smoldering risk area based on the occurrence time of each feature includes: Compare the appearance time of the smoke release characteristics and the accumulation characteristics. If the smoke release characteristics appear before the accumulation characteristics, it is determined that the analysis of the ignition risk area is incorrect.

5. The method for real-time fire scene prediction based on deep neural network according to claim 4 is characterized in that: After verifying the smoldering risk area based on the occurrence time of each feature, the location of the smoldering risk area is also verified: Determine the distance between each feature trigger position; If the distance between any two features is greater than the distance threshold, it is determined that the analysis of the smoldering risk area is wrong.

6. The method for real-time fire scene prediction based on deep neural network according to claim 1 is characterized in that: The degree parameters of each characteristic include the temperature and change gradient of the deposit, the smoke coverage area, the smoke rising speed and the accumulation range of the deposit.

7. The method for real-time prediction of fire scenes based on deep neural networks according to claim 6 is characterized in that: The real-time fire scene prediction method based on deep neural network also includes establishing a neural network model for predicting deflagration risk: Obtain training data from fire events, including the temperature of the deposit, the temperature gradient, the smoke coverage area, the smoke rising speed, the deposit accumulation range, and whether the deposit deflagrates and the values ​​of the training data corresponding to the deflagration time; Inputting the training data into the input layer of the neural network model; Accelerate model convergence and handle nonlinear features through ReLU; The sigmoid function is used to compress the output to the range of 0 to 1, which comprehensively represents the probability and triggering speed of the deflagration risk.

8. The method for real-time fire scene prediction based on deep neural network according to claim 1 is characterized in that: The deflagration risk levels used to determine smoldering risk include: Input the degree parameters of each characteristic of the smoldering risk area into the neural network model to obtain the output results; The deflagration risk level is divided according to the output results.

Images