A fire AI early warning system and method

The fire AI early warning system uses image sensors and deep learning neural network to judge the probability of fire, avoiding the fire extinguishing malfunction in traditional fire alarm systems, solving the problem of high false alarm rates, achieving safe fire extinguishing operations, and under cost constraints Get the best fire extinguishing effect.

CN118800045BActive Publication Date: 2024-11-22GUANGDONG ZHONGKE HUIJU TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411266447.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-11-22
Estimated Expiration
2044-09-11

AI Technical Summary

Technical Problem

Traditional fire alarm systems have false alarm rates due to sensor sensitivity and batch differences, resulting in malfunctions of fire extinguishing and secondary disasters.

Method used

A fire AI early warning system was designed, using image sensors and deep learning neural networks to judge the fire probability and the name of combustible substances through a central control device, and fire extinguishing instructions were issued only when the threshold was reached to avoid false alarms.

Benefits of technology

Effectively avoid fire extinguishing malfunctions, prevent secondary disasters, and obtain the best fire extinguishing plan under the cost constraints of nozzles or potassium oxide fire extinguishing agents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118800045B_ABST
    Figure CN118800045B_ABST
Patent Text Reader

Abstract

A fire AI early warning system and method, belonging to the technical field of alarm devices. In the system, each fire extinguishing subsystem obtains an image of the monitored area; the image is processed and sent to a central processing unit; the fire recognition model of the central processing unit determines the probability of fire and the name of the combustible material in the monitored area according to the image information provided by the image sensor, and the fire recognition model is trained by a CNN neural network; the comparison module is used to compare the probability of fire with the threshold, and when it is greater than or equal to the threshold, the image obtained by the image sensor in real time is sent to a remote terminal, and the comparison result is transmitted to the decision module; the decision module receives the fire extinguishing instruction sent by the remote terminal, and when the probability of fire provided by the fire recognition model is greater than or equal to the threshold and the instruction sent by the remote terminal is that fire extinguishing is required, the fire extinguishing instruction is sent to the regional fire extinguishing subsystem. The present invention will not cause fire extinguishing misoperation, thereby preventing secondary disasters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a fire AI early warning system and method, belonging to the technical field of alarm devices. Background Art

[0002] The traditional fire alarm system uses smoke and temperature sensors as the main sensors. Once an abnormality occurs in the protection area, the sound and light alarm will be activated. If there is a fire extinguishing system, the fire extinguishing can be activated in conjunction.

[0003] The disadvantage of the existing technology is that due to the sensitivity of the sensor and batch differences, there is a certain false alarm rate. Once the false alarm is linked, the fire extinguishing agent will be sprayed, causing secondary disasters. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the object of the present invention is to provide a fire AI early warning system and method, which will not cause false fire extinguishing operations and prevent secondary disasters.

[0005] To achieve the above-mentioned purpose, the present invention provides a fire AI early warning system, which includes: N regional fire extinguishing subsystems and a central control device, the central control device includes a central processor and a first communication unit, the n-th regional fire extinguishing subsystem includes an image sensor, a control unit and a second communication unit, the image sensor is used to obtain an image of the monitored n-th region; the second communication unit is connected to the first communication unit for communication, the control unit processes the image and sends it to the central processor through the second communication unit and the first communication unit in sequence, the central processor includes a fire recognition model, a comparison module and a decision module, the fire recognition model is based on the image transmission The image information provided by the sensor determines the probability of fire and the name of the combustible material in the monitored nth area; the comparison module is used to compare the probability of fire provided by the fire recognition model with the threshold value. When it is greater than or equal to the threshold value, the image acquired by the image sensor in real time is sent to the remote terminal through the first communication unit, and the comparison result is transmitted to the decision module; the decision module receives the instruction sent by the remote terminal. When the instruction sent by the remote terminal is to extinguish the fire and when the probability of fire provided by the fire recognition model is greater than or equal to the threshold value, the fire extinguishing instruction is sent to the fire extinguishing subsystem of the nth area, n=1,2,…,N, N is a positive integer greater than or equal to 2.

[0006] To achieve the above-mentioned purpose, the present invention also provides a fire AI early warning method, which comprises the following steps:

[0007] The n-area fire extinguishing subsystem is used to obtain an image of the monitored n-area through an image sensor, and processes the image through a control unit and then sends it to a central controller;

[0008] The central controller uses the fire recognition model to determine the probability of fire in the monitored area and the name of the combustible material based on the image information provided by the image sensor;

[0009] The central controller compares the probability of fire provided by the fire recognition model with the threshold through the comparison module. When it is greater than or equal to the threshold, the image acquired by the image sensor in real time is sent to the remote terminal, and the comparison result is transmitted to the decision module;

[0010] The central control receives the instructions sent by the remote terminal through the decision module. When the instruction sent by the remote terminal is to extinguish the fire and when the probability of fire provided by the fire identification model is greater than or equal to the threshold, the fire extinguishing instruction is sent to the fire extinguishing subsystem in the nth area, n=1,2,…,N, N is a positive integer greater than or equal to 2.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. No misoperation of fire extinguishing will occur, thus preventing secondary disasters; 2. The best fire extinguishing plan can be obtained under the nozzle of the sprinkler system or the fire extinguisher of potassium oxide fire extinguishing agent and cost constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a block diagram of the composition of the fire AI early warning system provided by the first embodiment of the present invention.

[0013] Figure 2 It is a block diagram of the composition of the training module provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0015] In the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0016] First embodiment

[0017] Figure 1 FIG. 1 is a block diagram of a fire AI early warning system provided by a first embodiment of the present invention. Figure 1As shown, the fire AI early warning system provided by the first embodiment of the present invention includes: N regional fire extinguishing subsystems and a central control device, the central control device includes a central processor and a first communication unit, the n-th regional fire extinguishing subsystem includes an image sensor, a control unit and a second communication unit, the image sensor is used to obtain an image of the monitored n-th region; the second communication unit is connected to the first communication unit for communication, the control unit processes the image and sends it to the central processor through the second communication unit and the first communication unit in sequence, the central processor includes a fire recognition model, a comparison module and a decision module, the fire recognition model determines the probability of fire and the name of the combustible material in the monitored n-th region according to the image information provided by the image sensor; the comparison module is used to compare the probability of fire provided by the fire recognition model with a threshold value, when it is greater than or equal to the threshold value, the image obtained by the image sensor in real time is sent to the remote terminal through the first communication unit, and the comparison result is transmitted to the decision module; the decision module receives the instruction sent by the remote terminal, when the instruction sent by the remote terminal is that fire extinguishing is required and when the probability of fire provided by the fire recognition model is greater than or equal to the threshold value, the fire extinguishing instruction is sent to the n-th regional fire extinguishing subsystem, n=1,2,…,N, N is a positive integer greater than or equal to 2.

[0018] The first embodiment of the present invention can achieve the beneficial effect of preventing false fire extinguishing operations and preventing secondary disasters through the above technical solution.

[0019] In the first embodiment, the central control device further comprises a first storage unit for storing computer programs and data. The computer programs, for example, comprise programs for implementing a fire identification model, a comparison module, a decision module and a training module.

[0020] The central control device also includes a display unit, which is used to display the monitoring images of each area and also to display a human-machine interface. A user can control each area according to the human-machine interface.

[0021] In the first embodiment, the first communication unit of the central control device and the second communication units of the N regional fire extinguishing subsystems communicate through the 5G network.

[0022] In the first embodiment, the n-th area fire extinguishing subsystem further includes a sprinkler system and a potassium oxide fire extinguishing agent fire extinguishing system; when the control unit receives a fire extinguishing instruction and the name of the combustible material, it activates one or both of the sprinkler system and the potassium oxide fire extinguishing agent fire extinguishing system according to the name of the combustible material to extinguish the fire. The fire extinguishing agent also includes a dry powder fire extinguishing agent, etc.

[0023] In the first embodiment, the control unit includes a deep learning neural network, the deep learning neural network includes a two-dimensional neuron layer, the two-dimensional neuron layer includes M×S neurons, and the M×S neurons respectively output the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing agent fire extinguishing system to the sth fire point ms (t); the spraying volume of the mth nozzle of the sprinkler system on the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system on the sth fire point ms (t) are fed back to the input ends of M×S neurons, where M is an integer greater than or equal to 1, m=1,...,M; M is the total number of nozzles or fire extinguishers; S is a positive integer greater than or equal to 1, s=1,...,S; S is the total number of fire points.

[0024] In the first embodiment, the deep learning neural network is a continuous Hopfield neural network, which optimizes the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system to the sth fire point. ms (t), the specific process includes:

[0025] S2-01: Use the following formula to calculate the activation equation of the neuron :

[0026] ,

[0027] In the formula, For the The input variables of each neuron; For the The output variable of each neuron; is the first The neuron and The strength of the connections between neurons; is the external input of the msth neuron under the minimum constraint; The availability of the external input nozzle or fire extinguisher for the ms-th neuron is Time The neuron and The strength of the connections between neurons; The external input cost for the ms-th neuron is Time The neuron and The interconnection strength between neurons; α0, α1 and α2 are learning rates; a=1,...,S; d=1,...,M;

[0028] S2-02: Use the first-order Euler method to calculate Input variables of neurons :

[0029] ;

[0030] S2-03: Calculate the Output variables of neurons :

[0031] ,

[0032] Where tansig(•) represents the sinusoidal double bending transfer function;

[0033] S2-04: Assign a value to t and calculate the minimum function according to the following formula:

[0034] ,

[0035] In the formula, is the value of the ignition point s; The probability that the mth nozzle or fire extinguisher implements the fire extinguishing plan on the sth fire point and successfully extinguishes the fire with a unit spraying amount or unit fire extinguishing volume; The plan for extinguishing the sth fire point with the mth nozzle or fire extinguisher; M is the total number of nozzles or fire extinguishers; S is the total number of fire points;

[0036] S2-05: Determine whether the minimum function E has reached the minimum. If the minimum function E has not reached the minimum value, return to step S2-01; if the minimum function E has reached the minimum value, the iteration ends and the output is .

[0037] The first embodiment of the present invention can achieve the beneficial effect of obtaining the best fire extinguishing plan under the constraints of nozzles or fire extinguishers and costs through the above technical solution.

[0038] In the first embodiment, the fire extinguishing subsystem of the nth area also includes a driving mechanism for carrying a camera for rotation. The central control device sends instructions to the control unit, and the control unit sends instructions to the driving mechanism according to the instructions, so that the driving mechanism drives the camera to rotate, thereby obtaining images of different perspectives of the monitored area.

[0039] In the first embodiment, the n-th zone fire extinguishing subsystem further includes a second storage unit for storing computer programs and data, the computer program being, for example, a program for implementing the functions of a deep learning neural network.

[0040] Figure 2 is a block diagram of the composition of the training module provided by the first embodiment of the present invention, such as Figure 2As shown, the fire recognition model is trained by a CNN neural network and a fully connected layer through a training module, and the training module includes: a data acquisition unit, a CNN neural network, a fully connected layer and an adjustment layer, wherein the data acquisition unit acquires an infrared image with a probability p of a fire of a combustible material contained in the monitored nth area, and the infrared image is acquired from an image acquired by an image sensor, and the infrared image is input into the CNN neural network as training data, and the probability p of a fire and the name of the combustible material are input into the adjustment layer as test data; the CNN neural network outputs the probability value of a fire and the name of the combustible material through the fully connected layer; the adjustment layer generates a loss function according to the probability value of a fire and the name of the combustible material output by the fully connected layer and the test data, calculates a gradient according to the loss function, and updates the parameters of the CNN neural network and the parameters of the fully connected layer according to the gradient.

[0041] In the first embodiment, the adjustment layer generates a loss function according to the probability value of fire occurrence, the name of the combustible material and the test data output by the fully connected layer, calculates the gradient according to the loss function, and updates the parameters of the CNN neural network and the parameters of the fully connected layer according to the gradient, specifically including:

[0042] S1-01: Get the output vector of the fully connected layer : , the output vector is the concatenated vector of the probability vector representing the occurrence of fire in the nth area and the name vector representing the combustible material.

[0043] In the formula, , f(•) represents a function;

[0044] Infrared image data as input to the CNN neural network; are the parameters of the CNN neural network; is the parameter of the fully connected layer, k=1,2,…,K;

[0045] S1-02: Get the vector of test data: , r nk is the kth value of the vector of the test data in the nth region, k=1,2,…,K; according to and The generated loss function L is:

[0046] , where log(•) represents the logarithmic function;

[0047] Calculate the parameters according to the following formula and parameters The gradient is:

[0048] ,

[0049] ;

[0050] S1-03: Update the current parameters of CNN according to the following formula for :

[0051] ;

[0052] Update the current parameters of the fully connected layer according to the following formula for :

[0053] ,

[0054] In the formula, ln[•] represents the natural logarithm function, , is the learning rate, Express The gradient of Express The gradient of

[0055] S1-04: Determine whether the loss function value is the minimum. If so, end the training and generate a fire recognition model. If not, make the new parameters Replace the current parameters , so that the new parameters Replace the current parameters And return to step S1-01.

[0056] The first embodiment of the present invention can achieve the goal of training the CNN neural network model into a fire recognition model that can recognize the probability of fire and the name of combustible materials through the above technical solution.

[0057] Second embodiment

[0058] The second embodiment of the present invention only describes the contents that are different from the first embodiment, and the same contents will not be described repeatedly.

[0059] The second embodiment of the present invention provides a fire AI early warning method, which includes the following steps:

[0060] The n-area fire extinguishing subsystem is used to obtain an image of the monitored n-area through an image sensor, and processes the image through a control unit and then sends it to a central controller;

[0061] The central controller uses the fire recognition model to determine the probability of fire in the monitored area and the name of the combustible material based on the image information provided by the image sensor;

[0062] The central controller compares the probability of fire provided by the fire recognition model with the threshold through the comparison module. When it is greater than or equal to the threshold, the image acquired by the image sensor in real time is sent to the remote terminal, and the comparison result is transmitted to the decision module;

[0063] The central controller receives instructions sent by the remote terminal through the decision module. When the instruction sent by the remote terminal is to extinguish the fire and when the probability of fire provided by the fire identification model is greater than or equal to the threshold, the fire extinguishing instruction is sent to the fire extinguishing subsystem in the nth area, n=1,2,…,N, N is a positive integer greater than or equal to 2.

[0064] The second embodiment of the present invention can achieve the beneficial effect of preventing false fire extinguishing operations and preventing secondary disasters through the above technical solution.

[0065] Third embodiment

[0066] The third embodiment of the present invention only describes the contents that are different from the second embodiment, and the same contents are not described again.

[0067] A third embodiment of the present invention provides a computer program product, which is a program code compiled in a computer language according to the method described in the second embodiment, and the program code can be called and executed by a processor.

[0068] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A fire AI early warning system, include: N regional fire extinguishing subsystems and a central control device, the central control device includes a central processor and a first communication unit, the nth regional fire extinguishing subsystem includes an image sensor, a control unit and a second communication unit, the image sensor is used to obtain an image of the monitored nth region; the second communication unit is connected to the first communication unit for communication, the control unit processes the image and sends it to the central processor through the second communication unit and the first communication unit in sequence, characterized in that the central processor includes a fire recognition model, a comparison module and a decision module, the fire recognition model determines the probability of fire and the name of the combustible material in the monitored nth region according to the image information provided by the image sensor; the comparison module is used to compare the probability of fire provided by the fire recognition model with a threshold value, when it is greater than or equal to the threshold value, the image obtained by the image sensor in real time is sent to the remote terminal through the first communication unit, and the comparison result is transmitted to the decision module; the decision module receives the instruction sent by the remote terminal, when the instruction sent by the remote terminal is that fire extinguishing is required and when the probability of fire provided by the fire recognition model is greater than or equal to the threshold value, the fire extinguishing instruction is sent to the nth regional fire extinguishing subsystem, n=1,2,…,N, N is a positive integer greater than or equal to 2; The control unit includes a deep learning neural network, the deep learning neural network includes a two-dimensional neuron layer, the two-dimensional neuron layer includes M×S neurons, and the M×S neurons respectively output the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing agent fire extinguishing system to the sth fire point ms (t); the spraying volume of the mth nozzle of the sprinkler system on the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system on the sth fire point ms (t) are fed back to the input ends of M×S neurons, where M is an integer greater than or equal to 1, m=1,...,M; M is the total number of nozzles or fire extinguishers; S is a positive integer greater than or equal to 1, s=1,...,S; S is the total number of fire points; The deep learning neural network is a continuous Hopfield neural network, which optimizes the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system to the sth fire point. ms (t), the specific process includes: S2-01: Use the following formula to calculate the activation equation of the neuron : , In the formula, For the The input variables of each neuron; For the The output variable of each neuron; is the first The neuron and The strength of the connections between neurons; is the external input of the ms-th neuron under the minimum function constraint; The availability of the external input nozzle or fire extinguisher for the ms-th neuron is Time The neuron and The strength of the connections between neurons; The cost of external input to the ms-th neuron is Time The neuron and The interconnection strength between neurons; α 0 , α 1 and α 2 is the learning rate; a=1,...,S; d=1,...,M; S2-02: Use the first-order Euler method to calculate Input variables of neurons : ; S2-03: Calculate the Output variables of neurons : , Where tansig(•) represents the sinusoidal double bending transfer function; S2-04: Assign a value to t and calculate the minimum function according to the following formula: , In the formula, is the value of the s-th ignition point; The probability that the mth nozzle or fire extinguisher implements the fire extinguishing plan on the sth fire point and successfully extinguishes the fire with a unit spraying amount or unit fire extinguishing volume; A plan for extinguishing the sth fire with the mth nozzle or extinguisher; S2-05: Determine whether the minimum function E has reached the minimum. If the minimum function E has not reached the minimum value, return to step S2-01; if the minimum function E has reached the minimum value, the iteration ends and the output is .

2. The fire AI early warning system according to claim 1, It is characterized in that The fire recognition model is trained by a CNN neural network and a fully connected layer through a training module. The training module includes: a data acquisition unit, a CNN neural network, a fully connected layer and an adjustment layer, wherein the data acquisition unit acquires an infrared image with a probability p of a fire of combustible materials contained in the monitored nth area, and the infrared image is acquired from an image acquired by an image sensor, and the infrared image data is input into the CNN neural network as training data, and the probability p of a fire and the name of the combustible material are input into the adjustment layer as test data; the CNN neural network outputs the probability value of a fire and the name of the combustible material through the fully connected layer; the adjustment layer generates a loss function according to the probability value of a fire and the name of the combustible material output by the fully connected layer and the test data, calculates a gradient according to the loss function, and updates the parameters of the CNN neural network and the parameters of the fully connected layer according to the gradient.

3. The fire AI early warning system according to claim 2, It is characterized in that The nth area fire extinguishing subsystem also includes a sprinkler system and a potassium oxide fire extinguishing agent fire extinguishing system; when the control unit receives a fire extinguishing instruction and the name of the combustible material, it starts one or both of the sprinkler system and the potassium oxide fire extinguishing agent fire extinguishing system according to the name of the combustible material to extinguish the fire.

4. A fire AI early warning method, It is characterized in that It includes the following steps: The n-area fire extinguishing subsystem is used to obtain an image of the monitored n-area through an image sensor, and processes the image through a control unit and then sends it to a central controller; The central controller uses the fire recognition model to determine the probability of fire in the monitored area and the name of the combustible material based on the image information provided by the image sensor; The central controller compares the probability of fire provided by the fire recognition model with the threshold through the comparison module. When it is greater than or equal to the threshold, the image acquired by the image sensor in real time is sent to the remote terminal, and the comparison result is transmitted to the decision module; The central controller receives the instructions sent by the remote terminal through the decision module. When the instruction sent by the remote terminal is to extinguish the fire and when the probability of fire provided by the fire identification model is greater than or equal to the threshold, the central controller sends a fire extinguishing instruction to the fire extinguishing subsystem in the nth area, n=1,2,…,N, where N is a positive integer greater than or equal to 2; The control unit includes a deep learning neural network, the deep learning neural network includes a two-dimensional neuron layer, the two-dimensional neuron layer includes M×S neurons, and the M×S neurons respectively output the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing agent fire extinguishing system to the sth fire point ms (t); the spraying volume of the mth nozzle of the sprinkler system on the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system on the sth fire point ms (t) are fed back to the input ends of M×S neurons, where M is an integer greater than or equal to 1, m=1,...,M; M is the total number of nozzles or fire extinguishers; S is a positive integer greater than or equal to 1, s=1,...,S; S is the total number of fire points; The deep learning neural network is a continuous Hopfield neural network, which optimizes the spraying amount of the mth nozzle of the sprinkler system to the sth fire point or the fire extinguishing volume V of the mth fire extinguisher of the potassium oxide fire extinguishing system to the sth fire point. ms (t), the specific process includes: S2-01: Use the following formula to calculate the activation equation of the neuron : , In the formula, For the The input variables of each neuron; For the The output variable of each neuron; is the first The neuron and The strength of the connections between neurons; is the external input of the ms-th neuron under the minimum function constraint; The availability of the external input nozzle or fire extinguisher for the ms-th neuron is Time The neuron and The strength of the connections between neurons; The external input cost for the ms-th neuron is Time The neuron and The interconnection strength between neurons; α 0 , α 1 and α 2 is the learning rate; a=1,...,S; d=1,...,M; S2-02: Use the first-order Euler method to calculate Input variables of neurons : ; S2-03: Calculate the Output variables of neurons : , Where tansig(•) represents the sinusoidal double bending transfer function; S2-04: Assign a value to t and calculate the minimum function according to the following formula: , In the formula, is the value of the s-th ignition point; The probability that the mth nozzle or fire extinguisher implements the fire extinguishing plan on the sth fire point and successfully extinguishes the fire with a unit spraying amount or unit fire extinguishing volume; The plan for extinguishing the sth fire point with the mth nozzle or fire extinguisher; M is the total number of nozzles or fire extinguishers; S is the total number of fire points; S2-05: Determine whether the minimum function E has reached the minimum. If the minimum function E has not reached the minimum value, return to step S2-01; if the minimum function E has reached the minimum value, the iteration ends and the output is .

5. The fire AI early warning method according to claim 4, It is characterized in that The fire recognition model is trained by a CNN neural network and a fully connected layer through a training module. The training module includes: a data acquisition unit, a CNN neural network, a fully connected layer and an adjustment layer, wherein the data acquisition unit acquires an infrared image with a probability p of a fire of combustible materials contained in the monitored nth area, and the infrared image is acquired from an image acquired by an image sensor, and the infrared image is input into the CNN neural network as training data, and the probability p of a fire and the name of the combustible material are input into the adjustment layer as test data; the CNN neural network outputs the probability value of a fire and the name of the combustible material through the fully connected layer; the adjustment layer generates a loss function according to the probability value of a fire and the name of the combustible material output by the fully connected layer and the test data, calculates a gradient according to the loss function, and updates the parameters of the CNN neural network and the parameters of the fully connected layer according to the gradient.

6. The fire AI early warning method according to claim 5, It is characterized in that The nth area fire extinguishing subsystem also includes a sprinkler system and a potassium oxide fire extinguishing agent fire extinguishing system; when the control unit receives a fire extinguishing instruction and the name of the combustible material, it starts one or both of the sprinkler system and the potassium oxide fire extinguishing agent fire extinguishing system according to the name of the combustible material to extinguish the fire.