Indoor fire early warning system and method based on multi-source sensing

Abstract

The invention belongs to the technical field of fire early warning, and discloses an indoor fire early warning system based on multi-source sensing, which comprises a multi-source sensor module, a feature extraction module, a decision module, an early warning module and a user interaction module, the multi-source sensor module realizes multi-dimensional fire characteristic data acquisition from four aspects of smoke, temperature, gas and infrared thermal imaging. By arranging the multi-source sensor module, a smoke sensor, a temperature sensor, a gas sensor and an infrared thermal imager are combined, multi-source monitoring of a fire is achieved, meanwhile, feature extraction and decision fusion are conducted on smoke concentration, temperature, gas concentration and an infrared thermal imaging picture through mutual cooperation of the feature extraction module and the decision module, and the accuracy of the fire monitoring is improved. Compared with a traditional single sensor, the system has the advantages of being higher in detection precision and low in false alarm rate, and fire early warning can be achieved in a real sense.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fire early warning, and particularly relates to an indoor fire early warning system and method based on multi-source sensing. BACKGROUND

[0002] Indoor fire refers to fire occurring in the interior space of a building, has the characteristics of rapid development, dense smoke, high temperature, and difficult evacuation, and poses a serious threat to the safety of personnel life and property, common causes include electrical failure, improper management of open flames, accumulation of flammable materials, human arson or accidents, and natural factors, and hazards include high-temperature burns, toxic smoke, oxygen deficiency asphyxia, building collapse, and secondary disasters.

[0003] In order to timely warn of indoor fires, it is necessary to use corresponding sensors to monitor key factors of fire, but existing fire detection techniques are single, mostly rely on single sensors, and single sensors are easily affected by the surrounding environment, so that the monitoring of fire is prone to misjudgment, causing false alarms of fire, and it is difficult to truly achieve early warning, and multiple false alarms can also cause panic among users, so improvement is needed. SUMMARY

[0004] The purpose of the present application is to provide an indoor fire early warning system and method based on multi-source sensing to solve the problems raised in the background.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: an indoor fire early warning system based on multi-source sensing, comprising: a multi-source sensor module, a feature extraction module, a decision module, a warning module, and a user interaction module, wherein,

[0006] The multi-source sensor module realizes multi-dimensional fire feature data collection from four aspects of smoke, temperature, gas, and infrared thermal imaging;

[0007] The feature extraction module is used for feature extraction of the multi-dimensional fire feature data collected by the multi-source sensor module;

[0008] The decision module evaluates fire risk according to the extracted multi-dimensional fire features;

[0009] The warning module is used for warning operation according to the evaluation result of the decision module;

[0010] The user interaction module is used for real-time display of data of each sensor in the multi-source sensor module, thermal imaging map, and historical alarm data, and can also modify threshold setting.

[0011] As a preferred technical solution of the present invention, the multi-source sensor module is divided into a smoke sensor unit, a temperature sensor unit, a gas sensor unit and an infrared thermal imager unit. The smoke sensor unit adopts a photoelectric scattering or ionization sensor, the temperature sensor unit adopts an NTC thermistor sensor, the gas sensor unit adopts an electrochemical sensor, and the infrared thermal imager unit adopts an uncooled focal plane detector.

[0012] As a preferred technical solution of the present invention, the feature extraction module is divided into a data preprocessing unit and a feature extraction unit. The data preprocessing unit performs noise reduction, calibration and time synchronization on the data collected by each sensor and infrared thermal imager in the multi-source sensor module, and the feature extraction unit is used to perform feature processing on the preprocessed data.

[0013] As a preferred technical solution of the present invention, the feature extraction unit is divided into two categories when performing feature extraction: time domain features and spatial domain features:

[0014] The time domain features include:

[0015] Temperature change rate, calculate the linear regression slope of the temperature in the last 10 seconds:

[0016]

[0017] Where, t i represents the time of the i-th sampling point; T i represents the temperature of the i-th sampling point; Represents the mean of timestamps within the sliding window; Indicates the mean temperature value in the sliding window; window size: 10 seconds, that is, 10 points from t to t-9;

[0018] The CO concentration rising rate is calculated by the first-order difference method for the CO concentration change at three consecutive sampling points:

[0019]

[0020] Where, [CO] t Indicates the CO concentration at the current moment; [CO] t-2 Indicates the CO concentration before two sampling points; Δt indicates the sampling interval;

[0021] The smoke concentration fluctuation index SVI is calculated as the ratio of the standard deviation of the smoke concentration to the mean:

[0022]

[0023] Where, [S] i represents the smoke concentration at the i-th sampling point; μ [S]Mean value of smoke concentration in sliding window; σ [S] Standard deviation of smoke concentration in sliding window; N represents window size, N = 10, i.e. 10 seconds;

[0024] The time domain features include:

[0025] High temperature area A hot Threshold segmentation is performed on the infrared thermal image, the number of high temperature pixels is counted and converted into physical area:

[0026]

[0027] In the formula, N hot The number of pixels in the infrared thermal image whose temperature exceeds the threshold value; A pixel The physical area corresponding to a single pixel is determined by camera calibration; D represents the distance between the infrared thermal imager and the target scene;

[0028] High temperature area center temperature T max The temperature average value of high temperature area pixels is obtained:

[0029]

[0030] In the formula, T i The temperature value of the i-th high temperature pixel in the infrared thermal image; I i The grayscale value of the i-th high temperature pixel; hot represents the set of high temperature pixels (pixels whose temperature exceeds the temperature threshold value);

[0031] Temperature distribution Shannon entropy H temp The Shannon entropy of the temperature histogram of the infrared thermal image is calculated, reflecting the degree of disorder of the temperature distribution:

[0032]

[0033] In the formula, p i The pixel proportion of the i-th temperature interval; N i The number of pixels in the i-th interval; N total The total number of pixels.

[0034] As a preferred technical solution of the present application, the decision module is divided into a fusion unit and an evaluation unit, the fusion unit calculates the probability of fire support of the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager according to the extracted features, and then generates a multi-sensor comprehensive confidence through the Dempster combination rule, and the specific calculation method is:

[0035] Smoke sensor:

[0036] Set fire-related concentration threshold according to historical data, convert current concentration value to probability by Sigmoid function;

[0037]

[0038] where C represents current concentration; C thres represents fire threshold; k represents slope parameter; m i (fire) smoke Probability of fire supported by ith smoke sensor;

[0039] Temperature sensor:

[0040] Set probability m i (fire) T of fire supported by ith temperature sensor under different temperature intervals

[0041] When T < 50℃, m i (fire) T = 0.1;

[0042] When 50℃ ≤ T < 80℃, m i (fire) T = 0.1;

[0043] When T > 80℃, m i (fire) T = 0.1;

[0044] Gas sensor:

[0045] Use logarithmic function to reflect nonlinear relationship between CO concentration and fire probability:

[0046]

[0047] where [CO] represents current concentration; [CO] base represents background concentration; λ represents decay coefficient; m i (fire) CO represents probability of fire supported by ith gas sensor;

[0048] Infrared thermal imager:

[0049] Take all high-temperature areas, calculate total area A hot , if A hot > 100 pixels, set P area = 0.8; otherwise, P area = 0.2,

[0050] Calculate probability of Shannon entropy when fire:

[0051]

[0052] Combining the high temperature area and the probability of Shannon entropy during the fire, different weights are given, and the formula is as follows:

[0053] m i (fire) hot =a·P area +b·P entropy

[0054] In the formula, a represents the weight of the high temperature area; b represents the weight of the Shannon entropy; P area represents the probability mapping of the high temperature area; P entropy represents the probability of the Shannon entropy during the fire; m i (fire) hot represents the probability of the i-th infrared thermal imager supporting the fire; H temp represents the Shannon entropy of the temperature histogram of the infrared thermal imaging image; H base represents the Shannon entropy value of the normal environment;

[0055] Therefore, the calculation formula of the multi-sensor comprehensive confidence CF is as follows:

[0056]

[0057] In the formula, m i (fire) represents the probability of the i-th sensor supporting the fire, which is divided into m i (fire) smoke , m i (fire) CO , m i (fire) T , and m i (fire) hot .

[0058] As a preferred technical solution of the present application, the evaluation unit is used to evaluate the fire risk level according to the multi-sensor comprehensive confidence obtained by the fusion unit, and the third level is set as CF≤0.7; the second level is set as 0.7<CF≤0.0.9; and the first level is set as CF>0.9 or the detection of the open fire.

[0059] As a preferred technical solution of the present application, the warning module performs hierarchical warning according to the risk level evaluated by the evaluation unit: when the fire risk level is the third level, no alarm action is needed; when the fire risk level is the second level, the second-level warning is performed; and when the fire risk level is the first level, the first-level warning is started.

[0060] As a preferred technical scheme of the present application, the user interaction module is divided into a display unit, an inquiry unit and a setting unit, the display unit is used for displaying real-time data of each sensor of the multi-source sensor module, the inquiry unit is used for user inquiry of historical alarm information, and the setting unit is used for user setting and modification of threshold values of each sensor.

[0061] In order to achieve the above-mentioned another object, the present application provides the following technical scheme: a method for early warning of indoor fire based on multi-source sensing, and the specific steps are as follows:

[0062] Step one: multi-source sensor arrangement

[0063] According to the division of the building structure, the monitoring area is divided, and then the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are arranged in the corresponding monitoring area, and the threshold values of the sensors are set;

[0064] Step two: real-time data monitoring

[0065] The smoke sensor, the temperature sensor and the gas sensor monitor the smoke concentration, temperature change and gas concentration in real time, obtain the corresponding data, and the infrared thermal imager can obtain the infrared thermal imaging image in real time;

[0066] Step three: data processing

[0067] The smoke concentration, temperature, gas concentration and infrared thermal imaging image collected by the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are preprocessed, and then the fire feature data is extracted;

[0068] Step four: risk assessment

[0069] According to the extracted fire feature data, decision fusion is carried out to obtain the fire probability, and then the risk level is evaluated according to the fire probability;

[0070] Step five: fire warning

[0071] According to the fire risk level, the user is warned of the fire;

[0072] Step six: processing

[0073] After receiving the fire alarm, the corresponding emergency measures are taken according to the alarm level, and the fire is handled in time.

[0074] The beneficial effects of the present application are as follows:

[0075] The present application realizes multi-source monitoring of fire by setting a multi-source sensor module, combining a smoke sensor, a temperature sensor, a gas sensor and an infrared thermal imager, and realizing feature extraction and decision fusion of smoke concentration, temperature, gas concentration and infrared thermal imaging by the cooperation of a feature extraction module and a decision module, thereby realizing multi-dimensional fire risk judgment and alarm. BRIEF DESCRIPTION OF DRAWINGS

[0076] Figure 1 The system block diagram of the present application is shown in the figure;

[0077] Figure 2 The system flow chart of the present application is shown in the figure. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0079] As shown in the figure, Figure 1 The present application provides an indoor fire early warning system based on multi-source sensing, comprising a multi-source sensor module, a feature extraction module, a decision module, a warning module and a user interaction module, wherein,

[0080] The multi-source sensor module realizes multi-dimensional fire feature data acquisition from four aspects of smoke, temperature, gas and infrared thermal imaging;

[0081] The feature extraction module is used for feature extraction of the multi-dimensional fire feature data collected by the multi-source sensor module;

[0082] The decision module evaluates fire risk according to the extracted multi-dimensional fire features;

[0083] The warning module is used for warning operation according to the evaluation result of the decision module;

[0084] The user interaction module is used for real-time display of the data of each sensor in the multi-source sensor module, the thermal imaging and the historical alarm data, and also can modify the threshold setting.

[0085] The cooperation of the multi-source sensor module and the feature extraction module can obtain effective fire information in time, the decision module can perform multi-dimensional fusion according to the data of the fire features, thereby facilitating accurate judgment of the fire probability, the early warning module realizes remote alarm, facilitating users to obtain alarm information in time, and the user interaction module facilitates the user to query the fire early warning and set the operation of each sensor threshold. Compared with a single sensor, multi-dimensional fire early warning is realized, thereby improving the accuracy of the early fire warning system.

[0086] The multi-source sensor module is divided into a smoke sensor unit, a temperature sensor unit, a gas sensor unit and an infrared thermal imager unit. The smoke sensor unit adopts a photoelectric scattering or ion sensor, the temperature sensor unit adopts an NTC thermistor sensor, the gas sensor unit adopts an electrochemical sensor, and the infrared thermal imager unit adopts a non-cooled focal plane detector.

[0087] The smoke sensor unit detects the concentration of suspended particles in the air through a photoelectric scattering or ion sensor, the temperature sensor unit monitors the ambient temperature and local hot spots in real time by deploying an NTC thermistor sensor, the gas sensor unit detects the CO concentration through an electrochemical sensor, and the infrared thermal imager unit generates a temperature distribution thermal map through a non-cooled focal plane detector.

[0088] The feature extraction module is divided into a data preprocessing unit and an extracted feature unit. The data preprocessing unit performs noise reduction, calibration and time synchronization on the data collected by each sensor and the infrared thermal imager in the multi-source sensor module, and the extracted feature unit is used for feature processing on the preprocessed data.

[0089] The specific processing of the data preprocessing unit is as follows: the smoke sensor eliminates random noise through Kalman filtering and smoothes instantaneous fluctuations through sliding average filtering, thereby suppressing the jitter of the smoke sensor and retaining the true concentration trend; the temperature sensor removes impulse noise through median filtering and corrects missing values caused by sensor failure through linear interpolation, thereby avoiding false judgment of fire due to temperature mutation; the gas sensor eliminates the interference of environmental factors on gas detection through zero-point calibration (exposed to pure air for 10 minutes every day) and correction of sensor drift through table lookup method; the infrared thermal imager eliminates the response difference of the detector through two-point correction method and performs bad point repair through neighborhood mean filling, thereby generating a high-quality temperature distribution thermal map. In addition, in the time synchronization, PTP protocol is adopted, and through the master-slave clock synchronization mechanism, the timestamp deviation of all sensors is controlled within ±1ms, time interpolation is performed on the thermal imager (frame rate 5Hz) and other sensors (1Hz), and a unified time axis data stream is generated.

[0090] In the feature extraction unit, the feature extraction is divided into time domain features and spatial domain features:

[0091] Time domain features include:

[0092] Temperature change rate, linear regression slope of temperature in the last 10 seconds:

[0093]

[0094] Where t i represents the time of the i-th sampling point; T i represents the temperature of the i-th sampling point; represents the mean of the time stamps in the sliding window; represents the mean of the temperature values in the sliding window; window size: 10 seconds, i.e. 10 points from t to t-9;

[0095] CO concentration rising rate, first-order difference method to calculate the CO concentration change of the last 3 sampling points:

[0096]

[0097] Where [CO] t represents the current CO concentration; [CO] t-2 represents the CO concentration 2 sampling points ago; Δt represents the sampling interval time;

[0098] Smoke concentration fluctuation index SVI, calculate the ratio of the standard deviation and the mean of the smoke concentration:

[0099]

[0100] Where [S] i represents the smoke concentration of the i-th sampling point; μ [S] represents the mean of the smoke concentration in the sliding window; σ [S] represents the standard deviation of the smoke concentration in the sliding window; N represents the window size, take N=10, i.e. 10 seconds;

[0101] Time domain features include:

[0102] High temperature area A hot , threshold segmentation of infrared thermal imaging image, statistics of high temperature pixel number and conversion to physical area:

[0103]

[0104] Where N hot represents the number of pixels in the infrared thermal imaging image whose temperature exceeds the threshold; A pixel represents the physical area corresponding to a single pixel, which is determined by camera calibration; D represents the distance between the infrared thermal imager and the target scene;

[0105] Center temperature of high temperature region T max The temperature average value of the high temperature region pixels is taken as:

[0106]

[0107] In the formula, T i represents the temperature value of the i-th high temperature pixel in the infrared thermal image; I i represents the gray value of the i-th high temperature pixel; hot represents a high temperature pixel set (pixels with a temperature exceeding a temperature threshold);

[0108] Temperature distribution Shannon entropy H temp The Shannon entropy of the temperature histogram of the infrared thermal image is calculated, reflecting the degree of disorder of the temperature distribution:

[0109]

[0110] In the formula, p i represents the pixel proportion of the i-th temperature interval; N i represents the number of pixels in the i-th interval; N total represents the total number of pixels.

[0111] Through feature extraction of multi-source sensors, effective information of fire occurrence can be obtained in time, so as to facilitate accurate judgment of fire.

[0112] Among them, the decision module is divided into a fusion unit and an evaluation unit, the fusion unit calculates the probability of supporting fire of the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager according to the extracted features, and then generates a comprehensive confidence of multi-sensor through the Dempster combination rule, and the specific calculation method is:

[0113] Smoke sensor:

[0114] According to historical data, set the concentration threshold related to fire, and convert the current concentration value into probability through the Sigmoid function;

[0115]

[0116] In the formula, C represents the current concentration; C thres represents the fire threshold; k represents the slope parameter; represents; m i (fire) smoke The probability of the i-th smoke sensor supporting fire;

[0117] Temperature sensor:

[0118] Set the probability m i (fire) T of the i-th temperature sensor supporting fire under different temperature intervals:

[0119] When T < 50℃, m i (fire) T = 0.1;

[0120] When 50℃≤T<80℃, m i (fire) T = 0.1;

[0121] When T > 80℃, m i (fire) T = 0.1;

[0122] Gas sensor:

[0123] The logarithmic function is used to reflect the nonlinear relationship between CO concentration and fire probability:

[0124]

[0125] In the formula, [CO] represents the current concentration; [CO] base represents the background concentration; λ represents the decay coefficient; m i (fire) CO represents the probability of the i-th gas sensor supporting fire;

[0126] Infrared thermal imager:

[0127] Take all high-temperature areas, calculate their total area A hot , if A hot > 100 pixels, set P area = 0.8; otherwise P area = 0.2,

[0128] Calculate the probability of Shannon entropy when fire:

[0129]

[0130] Combine the high-temperature area and the probability of Shannon entropy when fire, and give different weights, then:

[0131] m i (fire) hot = a·P area + b·P entropy

[0132] In the formula, a represents the weight of high-temperature area; b represents the weight of Shannon entropy; P area represents the probability mapping of high-temperature area; P entropy represents the probability of Shannon entropy when fire; m i (fire) hot represents the probability of the i-th infrared thermal imager supporting fire; H tempDenotes the Shannon entropy of the temperature histogram of the infrared thermal image; H base Denotes the Shannon entropy value for the normal environment;

[0133] In summary, the calculation formula for the multi-sensor comprehensive confidence CF is as follows:

[0134]

[0135] In the formula, m i (fire) represents the probability that the i-th sensor supports a fire, which is divided into m i (fire) smoke 、m i (fire) CO 、m i (fire) T and m i (fire) hot .

[0136] By calculating the multi-sensor comprehensive confidence through the fusion unit, the probability of a fire can be obtained, and thus the evaluation and judgment of the multi-dimensional fire probability are realized.

[0137] Among them, the evaluation unit is used to evaluate the fire risk level according to the multi-sensor comprehensive confidence obtained by the fusion unit. It is set that CF ≤ 0.7 is the third level; 0.7 < CF ≤ 0.9 is the second level; CF > 0.9 or detecting open fire is the first level.

[0138] By making a multi-level classification of the fire risk, it is convenient for the early fire warning system to carry out corresponding alarm processing in a timely manner.

[0139] Among them, the warning module conducts hierarchical alarms based on the risk level evaluated by the evaluation unit: when the fire risk level is the third level, no alarm action is required; when the fire risk level is the second level, a secondary alarm is carried out; when the fire risk level is the first level, a primary alarm is activated.

[0140] The alarm information of the warning module is pushed to the user interaction module through the MQTT protocol, which is convenient for timely reminding users to pay attention to the fire and take corresponding preventive measures.

[0141] Among them, the user interaction module is divided into a display unit, a query unit, and a setting unit. The display unit is used to display the real-time data of each sensor of the multi-source sensor module. The query unit is used for users to query historical alarm information. The setting unit is used for users to set and modify the thresholds of each sensor.

[0142] The user interaction module facilitates the user to learn the smoke concentration, temperature, gas concentration and infrared thermal imaging diagram in time, and also facilitates the user to learn the historical alarm record, and in addition, the threshold of each sensor can be modified according to the actual place and environment, and the accuracy of the fire early warning system is improved.

[0143] As shown in Figure 2 The embodiment of the application also provides a method for indoor fire early warning based on multi-source sensing, and the specific steps are as follows:

[0144] Step one: multi-source sensor arrangement

[0145] According to the building structure, the monitoring area is divided, and then the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are arranged in the corresponding monitoring area, and the threshold of each sensor is set;

[0146] Step two: real-time data monitoring

[0147] The smoke sensor, the temperature sensor and the gas sensor monitor the smoke concentration, temperature change and gas concentration in real time, acquire corresponding data, and the infrared thermal imager can acquire the infrared thermal imaging diagram in real time;

[0148] Step three: data processing

[0149] The smoke concentration, temperature, gas concentration and infrared thermal imaging diagram collected by the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are preprocessed, and then fire feature data extraction is acquired;

[0150] Step four: risk assessment

[0151] According to the extracted fire feature data, decision fusion is carried out, the fire probability is obtained, and then the risk level is evaluated according to the fire probability;

[0152] Step five: fire warning

[0153] According to the fire risk level, the user is warned of the fire;

[0154] Step six: processing

[0155] After receiving the fire alarm, corresponding emergency measures are taken according to the alarm level, and the fire is handled in time.

[0156] When arranging the multi-source sensor, the smoke sensor and the temperature sensor need to be installed on the ceiling (distance from the wall ≥0.5m), the gas sensor needs to be 1.5m away from the ground, the infrared thermal imager needs to cover the key area, and the visual angle overlap rate needs to be ≥30%.

[0157] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0158] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, and it is intended that the scope of the application be limited solely by the scope of the appended claims and the equivalents thereof.

Claims

1. A multi-source sensor based early warning system for indoor fire, characterized in that, It comprises a multi-source sensor module, a feature extraction module, a decision module, a pre-warning module and a user interaction module, wherein, The multi-source sensor module realizes multi-dimensional fire feature data collection from four aspects of smoke, temperature, gas and infrared thermal imaging; The feature extraction module is used for feature extraction on the multi-dimensional fire feature data collected by the multi-source sensor module; The decision module evaluates fire risk according to the extracted multi-dimensional fire features; The pre-warning module is used for pre-warning operation according to the evaluation result of the decision module; The user interaction module is used for real-time display of data of each sensor in the multi-source sensor module, thermal imaging map and historical alarm data, and also can modify threshold setting.

2. The multi-source sensor based indoor fire early warning system and method of claim 1, wherein: The multi-source sensor module is divided into a smoke sensor unit, a temperature sensor unit, a gas sensor unit and an infrared thermal imager unit, the smoke sensor unit adopts photoelectric scattering or ion sensor, the temperature sensor unit is NTC thermistor sensor, the gas sensor unit adopts electrochemical sensor, and the infrared thermal imager unit adopts non-refrigeration type focal plane detector.

3. The multi-source sensor based indoor fire early warning system and method of claim 1, wherein: The feature extraction module is divided into a data preprocessing unit and an extracted feature unit, the data preprocessing unit performs noise reduction, calibration and time synchronization on the data collected by each sensor in the multi-source sensor module and the infrared thermal imager, and the extracted feature unit is used for feature processing on the preprocessed data.

4. The multi-source sensor based indoor fire early warning system and method of claim 3, wherein: The extracted feature unit is divided into time domain feature and space domain feature when performing feature extraction: The time domain feature includes: Temperature change rate, linear regression slope of temperature in the last 10 seconds: where t i denotes the time of the i-th sample point; T i denotes the temperature of the i-th sample point; denotes the mean of the timestamps within the sliding window; denotes the mean of the temperature values within the sliding window; window size: 10 seconds, i.e. 10 points from t to t-9 CO concentration rising rate, CO concentration change amount of three continuous sampling points is calculated by first-order difference method: where [CO] represents the CO concentration at the current time point; [CO] represents the CO concentration at the time point 2 sampling intervals before the current time point; and Δt represents the sampling interval time. t where [CO] represents the CO concentration at the current time point; [CO] represents the CO concentration at the time point 2 sampling intervals before the current time point; and Δt represents the sampling interval time. t-2 where [CO] Smoke concentration fluctuation index SVI, ratio of smoke concentration standard deviation to mean value: Where, [S] i represents the smoke concentration at the i-th sampling point; μ [S] represents the mean value of smoke concentration in the sliding window; σ [S] represents the standard deviation of smoke concentration within the sliding window; N represents the window size, and N = 10 means 10 seconds; The space domain feature includes: High temperature area A hot Threshold segmentation was performed on the infrared thermographic image, and the number of high temperature pixels was counted and converted to physical area: In the formula, N hot represents the number of pixels in the infrared thermal imaging image whose temperature exceeds the threshold value; A pixel represents the physical area corresponding to a single pixel, determined by camera calibration; D represents the distance between the infrared thermal imager and the target scene; Center temperature T of high temperature region max Taking a temperature average value for the high temperature region pixels: In the formula, T i represents the temperature value of the i-th high-temperature pixel in the infrared thermal imaging image; I i represents the gray value of the i-th high-temperature pixel; hot represents a high-temperature pixel set (pixels with temperatures exceeding a temperature threshold value); Temperature distribution shannon entropy H temp The shannon entropy of the temperature histogram of the infrared thermal image is calculated, reflecting the degree of disorder of the temperature distribution: where p i represents the proportion of pixels in the i-th temperature interval; N i represents the number of pixels in the i-th interval; N total represents the total number of pixels.

5. The multi-source sensor based indoor fire early warning system and method of claim 1, wherein: The decision module is divided into a fusion unit and an evaluation unit, the fusion unit calculates the probability of supporting fire of the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager according to the extracted features, and then generates multi-sensor comprehensive confidence through Dempster combination rule, and the specific calculation method is as follows: Smoke sensor: According to historical data, set the concentration threshold related to fire, and convert the current concentration value into probability through Sigmoid function; where C denotes the current concentration; C thre s denotes the fire threshold; k denotes the slope parameter; denotes; m i (fire) smoke the probability that the ithsmoke sensor supports a fire; Temperature sensor: Set the probability m of the i-th temperature sensor supporting the fire in different temperature intervals i (fire) T : m = 0.1 when T < 50°C i (fire) T = 0.1; m = 0.1 when 50 °C < T < 80 °C i (fire) T = 0.1; m = 0 when T > 80°C i (fire) T = 0.1; Gas sensor: Logarithmic function is adopted to reflect the nonlinear relationship between CO concentration and fire probability: where [CO] represents the current concentration; [CO]bg represents the background concentration; λ represents the decay coefficient; m represents the slope of the linear regression line; and t represents the time. base where [CO] represents the current concentration; [CO]bg represents the background concentration; λ represents the decay coefficient; m represents the slope of the linear regression line; and t represents the time. i (fire) CO where [CO] represents the current concentration; [CO]bg represents the background concentration; λ represents the decay coefficient; m represents the slope of the linear regression line; and t Infrared thermal imager: Take all high temperature area, calculate its total area A hot , if A hot > 100 pixels, set P area = 0.8; otherwise P area = 0.2, Calculate the probability of Shannon entropy when fire: Combine the area of high temperature region and the probability of Shannon entropy when fire, give different weights, then: m i (fire) hot = a · P area + b · P entropy wherein a represents a high temperature area weight; b represents a Shannon entropy weight; P area represents a high temperature area probability map; P entropy represents a probability of Shannon entropy at fire; m i (fire) hot represents a probability of fire supported by the i-th infrared thermal imager; H temp represents a Shannon entropy of the infrared thermal image temperature histogram; H base represents a Shannon entropy value for normal environment; In summary, the calculation formula of multi-sensor comprehensive confidence CF is: where m i (fire) represents the probability that the ith sensor supports a fire, and is divided into m i (fire) smoke , m i (fire) CO , m i (fire) T and m i (fire) hot .

6. The multi-source sensor based indoor fire early warning system and method of claim 5, wherein: The evaluation unit is used for evaluating fire risk level according to the multi-sensor comprehensive confidence obtained by the fusion unit, sets CF≤0.7 as the third level, sets 0.7<CF≤0.0.9 as the second level, and sets CF>0.9 or detects visible fire as the first level.

7. The multi-source sensor based indoor fire early warning system and method of claim 1, wherein: The pre-warning module carries out graded alarm according to the risk level evaluated by the evaluation unit: when the fire risk level is the third level, no alarm action is needed; when the fire risk level is the second level, second-level alarm is carried out; when the fire risk level is the first level, first-level alarm is started.

8. The multi-source sensor based indoor fire early warning system and method of claim 1, wherein: The user interaction module is divided into a display unit, a query unit and a setting unit, the display unit is used for displaying real-time data of each sensor of the multi-source sensor module, the query unit is used for user query of historical alarm information, and the setting unit is used for user setting and modification of threshold values of each sensor.

9. A method for early warning of indoor fire based on multi-source sensing, characterized in that: The specific steps are as follows: Step one: multi-source sensor arrangement According to the building structure, the monitoring area is divided, then the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are arranged in the corresponding monitoring area, and the threshold values of the sensors are set; Step two: real-time data monitoring The smoke sensor, the temperature sensor and the gas sensor monitor the smoke concentration, temperature change and gas concentration in real time, obtain the corresponding data, and the infrared thermal imager can obtain the infrared thermal imaging image in real time; Step three: data processing The smoke concentration, temperature, gas concentration and infrared thermal imaging image collected by the smoke sensor, the temperature sensor, the gas sensor and the infrared thermal imager are preprocessed, and then the fire feature data is extracted; Step four: risk assessment According to the extracted fire feature data, decision fusion is carried out to obtain the fire probability, and then the risk level is evaluated according to the fire probability; Step five: fire pre-warning According to the fire risk level, the user is alarmed of the fire; Step six: processing After receiving the fire alarm, corresponding emergency measures are taken according to the alarm level, and the fire is handled in time.

Images