Fire detection device and method for identifying burning material

Abstract

To provide a fire detection device and a combustible material specification method that can more accurately detect whether a flame caused by combustion of a crib has occurred based on acoustic data collected in a monitoring area.SOLUTION: A fire detection device comprises: a microphone 10 that collects a sound generated in a monitoring area as acoustic data; and a computer 20 having a level transition calculation unit (spectrum analysis unit 21) that calculates transition over time of sound pressure level in a predetermined high frequency band from the acoustic data collected by the microphone, and a flame detection unit 22 that counts the number of times of change from a value less than a determination level, which is preset in relation to the transition over time of sound pressure level, to a value equal to or more than the determination level, and when determining that the number of times is equal to or more than a preset determination number of times, detects the occurrence of a flame caused by a specific combustible material that generates an abrupt sound during combustion, in the monitoring area.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a fire detection device and a method for identifying a burning object that detects a crib-caused flame in a monitored area based on acoustic data. [Background technology]

[0002] Various types of fire detectors are used to detect the occurrence of a fire in a monitored area, including differential spot detectors, constant temperature spot detectors, and photoelectric spot detectors.

[0003] There are also fire detection devices that detect the sounds that are generated when something burns during a fire (see, for example, Non-Patent Document 1). Non-Patent Document 1 discloses a method for conducting combustion experiments to capture the characteristics of sounds that occur during combustion, and for distinguishing between fire and normal conditions based on frequency analysis of the sounds, thereby determining whether a fire has occurred.

[0004] In particular, Non-Patent Document 1 describes that it is appropriate to specify a frequency range and identify the presence or absence of a combustion phenomenon based on the time change in the integral value of the overall power (level) rather than tracking a specific frequency and level. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] "Frequency Analysis of Combustion Sound (2nd Report)," Fire and Disaster Management Research Institute Bulletin No. 31 (1994) Summary of the Invention [Problem to be solved by the invention]

[0006] However, the conventional techniques have the following problems. In Non-Patent Document 1, the idea of ​​time change in the integral value of the power spectrum of sound is introduced as a measure to more reliably separate background noise from combustion noise.

[0007] Combustion sounds vary depending on the material being burned. The combustion sound of a crib, which is used as an ignition source, tends to have a smaller frequency spectrum in the low-frequency range than the combustion sound of heptane. With the conventional technology disclosed in Non-Patent Document 1, it was difficult to identify flames originating from a crib with such frequency characteristics.

[0008] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a fire detection device and a burning material identification method that can detect with higher accuracy whether or not a flame has occurred due to the burning of a crib, based on acoustic data collected in a monitored area. [Means for solving the problem]

[0009] The fire detection device according to the present disclosure includes a microphone that collects sounds generated in a monitoring area as acoustic data, and a detection unit that detects the presence of a sound source from the acoustic data collected by the microphone. Specified as a range of 3kHz to 10kHz a level transition calculation unit that calculates the time transition of the sound pressure level in a predetermined high frequency band; and a counting unit that counts the number of times that the time transition of the sound pressure level changes from a value less than a predetermined judgment level to a value equal to or greater than the judgment level, and when the number of times is equal to or greater than the predetermined judgment number, If yes In the monitoring area, a fire occurred due to a specific combustible material that produced a sudden sound when burning. It is determined that It is equipped with a flame detection unit.

[0010] The fire detection device according to the present disclosure also includes a microphone that collects sounds generated in a monitoring area as acoustic data, and a detector that detects the presence of a sound from the acoustic data collected by the microphone. Specified as a range of 3kHz to 10kHz a level transition calculation unit that calculates the time transition of the sound pressure level in a predetermined high frequency band; and a unit that counts the number of times the sound pressure level changes from a value less than a predetermined judgment level to a value equal to or greater than the judgment level with respect to the time transition of the sound pressure level, and Certain combustible materials produce sudden noises when burning A feature calculation unit outputs the feature used to determine the occurrence of flames, and a sudden sound generated during combustion is occurs more frequently at moisture contents of 24% or more The system includes a model creation unit that creates a machine learning model for determining whether a flame has been generated by a specific combustible material by using, as parameters, feature quantities calculated by the feature calculation unit for each of various acoustic data collected when a flame, which is the target of detection, is generated in advance by burning a specific combustible material, and various acoustic data collected when no flame has been generated, including factors that may cause a false alarm; and a flame detection unit that, when monitoring a monitored area, uses the feature quantities calculated by the feature calculation unit as input to the machine learning model and determines whether a flame has been generated by a specific combustible material.

[0011] The method for identifying a combustible material according to the present disclosure includes a sound collection step of collecting sounds generated in a monitoring area as acoustic data using a microphone, and Specified as a range of 3kHz to 10kHz a level transition calculation step of calculating a time transition of a sound pressure level in a predetermined high frequency band; and a step of counting the number of times that the time transition of the sound pressure level changes from a value less than a predetermined judgment level to a value equal to or greater than the judgment level, and if the number of times is equal to or greater than the predetermined judgment number, If yes In the surveillance area , when burning The sudden sound caused by certain burning materials When a fire broke out and a specific step of specifying the [Effects of the Invention]

[0012] According to the present disclosure, a fire detection device and a burning material identification method can be obtained that can detect with higher accuracy whether a flame has occurred due to the burning of a crib based on acoustic data collected in a monitored area. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is an explanatory diagram showing the overall configuration of a fire detection device according to a first embodiment of the present disclosure. [Figure 2]FIG. 2 is an explanatory diagram comparing the difference in frequency characteristics between the combustion noise caused by heptane and the combustion noise caused by a crib in the first embodiment of the present disclosure. [Figure 3] FIG. 2 is an explanatory diagram showing a comparison of frequency spectra for three different types of cribs according to the first embodiment of the present disclosure. [Figure 4] FIG. 1 is an explanatory diagram showing a comparison of spectrograms and sound pressure time waveforms for three different types of crib according to the first embodiment of the present disclosure. [Figure 5] FIG. 2 is an explanatory diagram comparing the difference in the transition of sound pressure levels calculated from spectrograms of the combustion noise of heptane and the combustion noise of a crib in the first embodiment of the present disclosure. [Figure 6] FIG. 10 is an explanatory diagram comparing the difference in the number of times sudden sounds occur between combustion sounds caused by heptane and combustion sounds caused by a crib in the first embodiment of the present disclosure. [Figure 7] FIG. 2 is an explanatory diagram comparing differences in frequency characteristics between the combustion noise of heptane, the combustion noise of a crib, and the noise of a ventilation fan in the first embodiment of the present disclosure. [Figure 8] FIG. 10 is an explanatory diagram comparing the differences in the transition of sound pressure levels calculated from spectrograms of the combustion noise caused by a crib, the noise caused by a ventilation fan, and the combustion noise caused by heptane in the first embodiment of the present disclosure. [Figure 9] FIG. 10 is an explanatory diagram comparing the difference in the number of times sudden sounds occur among the combustion sound of heptane, the combustion sound of a crib, and the sound of a ventilation fan in the first embodiment of the present disclosure. [Figure 10] 3 is a flowchart showing a series of processes related to a fire detection method executed in the fire detection device according to the first embodiment of the present disclosure. [Figure 11] FIG. 10 is an explanatory diagram showing the overall configuration of a fire detection device according to a second embodiment of the present disclosure. [Figure 12] 10 is a flowchart showing a series of processes relating to a fire detection method executed in a fire detection device according to a second embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, preferred embodiments of the fire detection device and the method for identifying a burning object according to the present disclosure will be described with reference to the drawings. The present disclosure has a technical feature in that it extracts features specific to flames caused by crib combustion from the results of frequency analysis based on acoustic data collected in the monitored area and determines whether or not a flame has occurred, thereby enabling highly accurate detection of crib combustion, which was difficult to identify using conventional technology.

[0015] In this disclosure, "crib" refers to a general term for a specific combustible material that produces a sudden noise when burned.

[0016] Embodiment 1 1 is an explanatory diagram showing the overall configuration of a fire detection device according to a first embodiment of the present disclosure. The fire detection device according to the first embodiment includes a microphone 10 and a computer 20.

[0017] The microphone 10 is a sound collection device that is installed in the monitoring area of ​​the flame and collects sounds generated in the monitoring area as acoustic data. Note that it is also possible to install a plurality of microphones 10 in the monitoring area.

[0018] When determining whether or not there is a fire based on the acoustic data collected by the microphone 10, the focus is on the acoustic data generated by the combustion itself, rather than monitoring the smoke generated as a by-product of the fire, allowing for faster fire detection. Furthermore, the microphone 10 does not need to be installed on the ceiling like a smoke detector, allowing for greater freedom in determining the installation location.

[0019] The computer 20 is a controller that performs arithmetic processing on the acoustic data collected by the microphone 10 to determine whether a crib has caused a fire in the monitored area, and is equipped with a spectrum analysis unit 21 and a flame detection unit 22.

[0020] The spectrum analysis unit 21 calculates a spectrogram by performing frequency analysis on the acoustic data collected by the microphone 10. The flame detection unit 22 determines whether or not the spectrogram calculated by the spectrum analysis unit 21 contains a characteristic feature resulting from the burning of the crib. Furthermore, if the flame detection unit 22 determines that the characteristic feature is contained, it detects that a flame has broken out in the monitored area due to the burning of the crib.

[0021] 2 to 9, specific processing contents by spectrum analysis unit 21 and flame detection unit 22 will be described in detail. Fig. 2 is an explanatory diagram comparing the difference in frequency characteristics between the combustion sound caused by heptane and the combustion sound caused by a crib in the first embodiment of the present disclosure.

[0022] Figure 2(A) shows the frequency spectrum for the heptane combustion noise and the crib combustion noise. Comparing the two, the crib combustion noise tends to have a lower frequency spectrum level in the low frequency range compared to the heptane combustion noise, and conversely, tends to have a higher frequency spectrum level in the high frequency range.

[0023] Figure 2(B) is a spectrogram of the sound of heptane burning, and Figure 2(C) is a spectrogram of the sound of a crib burning. The spectrograms shown in Figures 2(B) and (C) show the time-dependent change in frequency components as a continuous brightness, with the vertical axis representing frequency on a logarithmic scale and the horizontal axis representing time.

[0024] Comparing Figure 2(B) with Figure 2(C), the spectrogram of the burning sound of the crib shows sudden crib-specific sounds (corresponding to popping sounds) occurring in the frequency range above approximately 3000 Hz. Therefore, in this disclosure, we focus on the occurrence of sudden sounds in this high-frequency range to detect crib combustion with high accuracy.

[0025] 3 is an explanatory diagram showing a comparison of frequency spectra for three different types of cribs according to the first embodiment of the present disclosure. Specifically, the diagram shows the results of calculating frequency spectra for three types of cedar cribs with moisture contents of 6%, 12%, and 24%.

[0026] Comparing the three frequency spectra, there is little difference in level in the low frequency range below 100 Hz, but differences are observed in the high frequency range, with the higher the moisture content, the higher the level.

[0027] 4 is an explanatory diagram showing a comparison of spectrograms and sound pressure time waveforms for three different types of cribs according to the first embodiment of the present disclosure. Specifically, similar to FIG. 3, the diagram shows the results of calculating spectrograms and sound pressure time waveforms for three types of cedar cribs with moisture contents of 6%, 12%, and 24%.

[0028] In FIG. 4, (A1), (A2), (B1), (B2), (C1), and (C2) respectively indicate the following: (A1): Time waveform of sound pressure of the burning sound of a cedar crib with a moisture content of 24% (A2): Spectrogram of the burning sound of a cedar crib with a moisture content of 24% (B1): Time waveform of sound pressure of the burning sound of a cedar crib with a moisture content of 12% (B2): Spectrogram of the burning sound of a cedar crib with a moisture content of 12% (C1): Time waveform of sound pressure of the burning sound of a cedar crib with a moisture content of 6% (C2): Spectrogram of the burning sound of a cedar crib with a moisture content of 6%

[0029] Here, the time waveforms shown in (A1), (B1), and (C1) correspond to time-series data of sound pressure, with the vertical axis representing sound pressure [Pa] and the horizontal axis representing time [s]. Also, the spectrograms shown in (A2), (B2), and (C2) represent the time-dependent changes in frequency components as a continuous brightness, with the vertical axis representing frequency on a logarithmic scale and the horizontal axis representing time.

[0030] 4, it can be seen that the higher the moisture content, the more frequent the sudden sounds that occur. Therefore, in the first embodiment, attention is focused on this difference, and the combustion noise caused by the crib is quantitatively identified by determining the transition of the sound pressure level over time in the high frequency range.

[0031] 5 is an explanatory diagram comparing the difference in the transition of sound pressure levels calculated from spectrograms of the combustion noise of heptane and the combustion noise of the crib in the first embodiment of the present disclosure. (A), (B), and (C) in FIG. 5 respectively indicate the following: (A): Spectrogram of the combustion sound of heptane (B): Spectrogram of the burning sound of the crib, corresponding to the cedar crib with a moisture content of 24% in Figure 4(A2). (C): Sound pressure level transition in the frequency band from 3 kHz to 10 kHz calculated from each spectrogram

[0032] For example, when f1=3000 [Hz], f2=10000 [Hz], and n is the number of samples from f1 to f2, the transition of the sound pressure level in the frequency band from 3 kHz to 10 kHz can be calculated by the following formula (1).

[0033]

number

[0034] Furthermore, normalization by the average value can be performed using the following equation (2), and FIG. 5(C) shows the level transition after normalization.

[0035]

number

[0036] In addition to extracting the level transition from the spectrogram, it can also be handled by limiting the band using a band-pass filter or high-pass filter to pass only the desired high-frequency band, and then extracting the time-series waveform of the sound pressure.

[0037] As shown in Figure 5(C), for example, when the judgment level is set to 15 dB, if we compare the number of times the value changes from below the judgment level to above the judgment level, the number of times the crib combustion sound is counted is greater than the number of times the heptane combustion sound is counted.

[0038] The threshold level can be set in advance based on the average background noise level in the monitored area when no fire is occurring. If the background noise is high, the threshold level can be set to a lower level (for example, 6 dB higher than the average background noise), and if the background noise is low, the threshold level can be set to a higher level (for example, 15 dB higher than the average background noise).

[0039] Figure 6 is an explanatory diagram comparing the difference in the number of sudden noise occurrences between heptane combustion sounds and crib combustion sounds in the first embodiment of the present disclosure. Specifically, the graph shows the count results of the number of times the level transition shown in Figure 5(C) changes from a value below the determination level to a value equal to or greater than the determination level. Therefore, by setting the determination count to an appropriate value, it is possible to distinguish between heptane combustion sounds and crib combustion sounds.

[0040] The above describes a method for distinguishing between the burning sounds of heptane and the burning sounds of a crib. Next, we will explain how this method can also be used to distinguish between the burning sounds of a ventilation fan, which is one of the causes of false alarms, and the burning sounds of a crib.

[0041] FIG. 7 is an explanatory diagram comparing the differences in frequency characteristics between the burning sound of heptane, the burning sound of a crib, and the sound of a ventilation fan in the first embodiment of the present disclosure. As shown in FIG. 7, the frequency spectrum of the burning sound of a crib is difficult to detect because it lacks low-frequency power. Furthermore, using only the frequency spectrum, the sound of a ventilation fan also contains high-frequency sounds, making it difficult to distinguish it from the crib, which can lead to false detection.

[0042] Therefore, as described above, by focusing on the level transition in the high frequency range and detecting the sudden sounds specific to the burning sound of the crib as a feature, it becomes possible to distinguish between the ventilation fan and the crib.

[0043] FIG. 8 is an explanatory diagram comparing the differences in the transitions in sound pressure levels calculated from spectrograms of the burning noise caused by the crib, the sound caused by the ventilation fan, and the burning noise caused by heptane in the first embodiment of the present disclosure.

[0044] In FIG. 8, (A1), (A2), (B1), (B2), (C1), and (C2) respectively indicate the following contents. (A1): Spectrogram of the burning sound of a cedar crib with a moisture content of 24% (A2): Level transition of the burning noise of a cedar crib with a moisture content of 24% (B1): Spectrogram of the sound of a ventilation fan (B2): Level transition of ventilation fan noise (C1): Spectrogram of the combustion sound of heptane (C2): Level transition of heptane combustion noise

[0045] As shown in (A2), (B2), and (C2) of Figure 8, for example, when the judgment level is set to 15 dB, comparing the number of times the value changes from below the judgment level to above the judgment level, the sound of the ventilation fan shows a transition similar to that of the heptane sound, and can be distinguished from the sound of the crib burning.

[0046] 9 is an explanatory diagram comparing the difference in the number of sudden noise occurrences for heptane combustion noise, crib combustion noise, and ventilation fan noise in the first embodiment of the present disclosure. (A) and (B) in FIG. 9 respectively indicate the following: (A): A combined image showing the level transitions of the heptane combustion noise, the crib combustion noise, and the ventilation fan noise. (B): Number of sudden sound occurrences in the high frequency range (3 kHz to 10 kHz) for heptane combustion, crib combustion, and ventilation fan noise.

[0047] As shown in FIG. 9(B), by setting the number of judgments to an appropriate value, the sound of the crib burning can be distinguished from the sound of heptane burning and the sound of the ventilation fan.

[0048] Next, a series of processes executed in the fire detection device according to the embodiment 1 will be described using a flowchart. Fig. 10 is a flowchart showing a series of processes related to a fire detection method executed in the fire detection device according to the embodiment 1 of the present disclosure.

[0049] First, in step S1001, the spectrum analysis unit 21 performs a collection process of sound data generated within the monitoring area via the microphone 10 installed in the monitoring area. Next, in step S1002, the spectrum analysis unit 21 performs a spectrum analysis process on the sound data collected by the microphone 10 to calculate a spectrogram.

[0050] Next, in step S1003, the flame detection unit 22 executes a calculation process for the transition of sound pressure level over time in a predetermined high frequency band. In the following description, the transition of sound pressure level over time will be simply referred to as "level transition over time" or "sound pressure level transition over time." The flame detection unit 22 can calculate the level transition over time by using the above formula (1).

[0051] Next, in step S1004, the flame detection unit 22 counts the number of times the level changes over time from a value below a preset judgment level to a value above the judgment level across the high frequency band. As an example of the judgment level, 15 dB can be used, as shown in FIG. 9.

[0052] Next, in step S1005, flame detection unit 22 determines whether the number of times counted in step S1004 is equal to or greater than a predetermined determination number. If the number of times counted is equal to or greater than the determination number, flame detection unit 22 determines that a flame caused by the crib has occurred, and proceeds to step S1006. On the other hand, if the number of times counted is not equal to or greater than the determination number, flame detection unit 22 determines that a flame caused by the crib has not occurred, and ends the series of processes.

[0053] If the process proceeds to step S1006, the flame detection unit 22 identifies the burning object as a crib and executes a flame detection and notification process. For example, the flame detection unit 22 can activate a fire extinguishing system, activate the linked operation of smoke prevention and exhaust devices such as shutters, and transfer a report to an emergency broadcast system, etc. by transmitting a fire signal to a fire receiver notifying the detection of a flame caused by a crib.

[0054] It should be noted that step S1001 in FIG. 10 corresponds to a sound collection step, steps S1002 and S1003 correspond to a level transition calculation step, and steps S1004 and S1006 correspond to a specification step.

[0055] As described above, according to the first embodiment, it is possible to calculate the characteristic amount of a crib flame from the frequency analysis results based on the sound data collected in the monitoring area, and quantitatively determine whether a crib flame has occurred in the monitoring area. Specifically, by calculating the characteristic amount corresponding to the frequency of sudden sound occurrence from the time transition of the sound pressure level in a predetermined high frequency band, it is possible to identify whether the burning object is a crib and detect whether a flame has occurred with high accuracy by distinguishing it from the cause of a false alarm.

[0056] In particular, the fire detection device of this embodiment 1 has a technical feature in that it determines whether a flame has occurred by performing calculations based on acoustic data collected within the monitored area, and provides the following effects.

[0057] Effect 1: The microphones can be installed in any position, allowing for easy retrofitting or temporary addition of fire monitoring functionality. In other words, a relatively simple configuration can be used to achieve fire detection processing in various monitoring areas.

[0058] Effect 2: Rather than monitoring smoke that is generated as a by-product of a fire, the fire detection process focuses on the acoustic data generated by the combustion itself, enabling faster fire detection.

[0059] Effect 3: By combining the fire detection method of the present disclosure with fire detection methods using other sensors, it is possible to suppress false alarm factors and more accurately detect whether a flame has been generated by a crib. For example, when a possible fire is detected using another fire detection method, the fire detection method of the present disclosure can be further applied to identify whether the sound is the burning sound of a crib.

[0060] Embodiment 2 In the first embodiment, the frequency of sudden sounds in the high frequency band is calculated as a feature from the results of frequency analysis of the acoustic data, and when the feature reaches a predetermined number of determinations, the burning object is identified as a crib and a crib-caused flame is detected. In contrast, in the second embodiment, the calculated feature is used as an input for a learning model to determine whether a crib-caused flame has occurred.

[0061] 11 is an explanatory diagram showing the overall configuration of a fire detection device according to Embodiment 2 of the present disclosure. The fire detection device according to Embodiment 2 includes a microphone 10 and a computer 20.

[0062] The microphone 10 is installed in the flame monitoring area and collects sounds generated within the monitoring area as acoustic data. The computer 20 is a controller that performs arithmetic processing on the acoustic data collected by the microphone 10 to determine whether a crib-caused flame has occurred in the monitoring area, and includes a spectrum analysis unit 21 and a flame detection unit 22.

[0063] Comparing the configuration of Fig. 11 in the present embodiment 2 with the configuration of Fig. 1 in the previous embodiment 1, the difference in the present embodiment 2 is that flame detection unit 22 has, as its internal components, feature amount calculation unit 23 and support vector machine 24. Therefore, the following description will focus on the differences between embodiment 1 and embodiment 2.

[0064] In the first embodiment described above, the flame detection unit 22 performs all of the following functions 1 to 3. Function 1: A function for calculating the time transition of sound pressure levels for a predetermined high frequency band from the results of the spectrum analysis by the spectrum analysis unit 21. Function 2: A function that performs counting using a judgment level regarding the time transition of sound pressure level, and calculates the number of times the level exceeds the judgment level as a feature. Function 3: A function that compares the feature value with a preset number of judgments, and if the feature value is equal to or greater than the number of judgments, identifies the burning object as a crib and outputs a judgment result that a flame caused by a crib has occurred in the monitored area.

[0065] In contrast to this, in the present embodiment 2, the part up to the calculation of the feature amounts by Function 1 and Function 2 is performed by the feature amount calculation unit 23. Furthermore, in the support vector machine 24 in the present embodiment 2, a learning model using the feature amounts as input parameters is created in advance, and the feature amounts calculated by the feature amount calculation unit 23 are input to the learning model, and whether or not a flame has been generated by the crib is determined by machine learning.

[0066] That is, in the second embodiment, instead of setting the number of determinations in advance, a learning model is created in which the feature values ​​are used as input parameters to determine whether a crib has caused a fire. Therefore, the support vector machine 24 will be further explained below.

[0067] The support vector machine 24 is one of the identification methods using supervised learning, and is a known learning model. In the second embodiment, in order to identify whether or not a flame is occurring due to the crib, a learning model is created based on the calculation results of feature amounts in various situations.

[0068] That is, a learning model is created in advance using the support vector machine 24, using as input parameters the features based on the acoustic data when a flame caused by the crib, which is the detection target, actually occurs, and the features based on the acoustic data when no flame occurs, including factors that may cause a false alarm.

[0069] As a result, the support vector machine 24 in the flame detection unit 22 can use the features calculated by the feature calculation unit 23 as input and use machine learning to determine whether or not a crib-caused flame is occurring within the monitored area.

[0070] By having such a configuration, the fire detection device of this embodiment 2 creates a learning model in advance and applies the features calculated from the acoustic data collected during monitoring as input parameters to the trained support vector machine 24, thereby being able to more accurately detect whether a crib has caused a fire within the monitoring area and distinguish it from false alarm factors.

[0071] Next, a series of processes executed in the fire detection device according to the second embodiment will be described using a flowchart. Fig. 12 is a flowchart showing a series of processes related to a fire detection method executed in the fire detection device according to the second embodiment of the present disclosure.

[0072] First, in step S1201, the spectrum analysis unit 21 performs a collection process of sound data generated within the monitoring area via the microphone 10 installed in the monitoring area. Next, in step S1202, the spectrum analysis unit 21 performs a spectrum analysis process on the sound data collected by the microphone 10 to calculate a spectrogram.

[0073] Next, in step S1203, the flame detection unit 22 executes a calculation process for the transition of sound pressure level over time in a predetermined high frequency band. In the following description, the transition of sound pressure level over time will be simply referred to as "level transition over time" or "sound pressure level transition over time." The flame detection unit 22 can calculate the level transition over time by using the above formula (1).

[0074] Next, in step S1204, the flame detection unit 22 counts the number of times the level transition over time exceeds the judgment level. Specifically, the flame detection unit 22 counts the number of times the level transition over time changes from a value below a preset judgment level to a value above the judgment level over the high frequency range, and obtains this number as a feature. As an example of the judgment level, 15 dB can be used, as shown in FIG. 9.

[0075] Next, in step S1205, the support vector machine 24 creates a learning model in advance, and executes flame detection processing using the machine learning model, using the number of times calculated as the feature amount in step S1204 as an input.

[0076] Next, in step S1206, if the support vector machine 24 determines that a flame has been generated by the crib as a result of executing the flame detection process using the machine learning model, the process proceeds to step S1207.

[0077] On the other hand, if the support vector machine 24 determines, as a result of executing the flame detection process using the machine learning model, that a flame has not been generated by the crib, it ends the series of processes.

[0078] If the process proceeds to step S1207, the flame detection unit 22 executes a process to identify the burning object as a crib and also executes a flame detection and notification process. For example, the flame detection unit 22 can activate a fire extinguishing system, activate the linked operation of smoke prevention and exhaust devices such as shutters, transfer a report to an emergency broadcast system, etc. by transmitting a fire signal to the fire receiver notifying that a flame caused by a crib has been detected.

[0079] As described above, according to the second embodiment, it is possible to calculate the characteristic features of a crib flame from the frequency analysis results based on the sound data collected in the monitored area, and quantitatively determine whether a crib flame has occurred in the monitored area using a pre-trained machine learning model. Specifically, by calculating the characteristic features corresponding to the frequency of sudden sound occurrences from the time transition of the sound pressure level in a predetermined high frequency band, it is possible to identify whether the burning object is a crib and detect whether a flame has occurred with high accuracy by distinguishing it from false alarm factors.

[0080] In other words, instead of setting the number of judgments in advance, by creating a learning model, it is possible to calculate a feature value corresponding to the frequency of sudden sound occurrence from the time progression of sound pressure levels in a specified high frequency band, thereby making it possible to detect with high accuracy whether a flame has been generated by the crib, and to achieve effects 1 to 3, as in the previous embodiment 1.

[0081] In addition, in the previous Figure 11, a configuration was shown in which the feature calculation unit 23 and the support vector machine 24 are included in the flame detection unit 22, but it is also possible to configure the feature calculation unit 23 and the support vector machine 24 independent of the flame detection unit 22.

[0082] In this case, the feature calculation unit specializes in calculating the features, the support vector machine 24 specializes in pre-generating the machine learning model, and the flame detection unit 22 specializes in determining whether a flame has been generated by the crib when monitoring the monitoring area, by using the features calculated by the feature calculation unit 23 as input parameters for the machine learning model pre-generated by the support vector machine 24.

[0083] Furthermore, in the first and second embodiments, a case has been described in which a spectrogram is obtained from acoustic data and then the time transition of the sound pressure level in the high frequency band is obtained. However, it is also possible to obtain the time transition of the sound pressure level in the high frequency band by performing an appropriate filtering process on the acoustic data.

[0084] In other words, a spectrum analysis unit that performs spectrum analysis processing is not a required component, and it is sufficient to have a configuration that includes a level transition calculation unit that can calculate the time transition of sound pressure levels in the high band based on acoustic data.

[0085] In the second embodiment, a support vector machine is used as the model creation unit that creates the machine learning model, but the model creation unit is not limited to this. The same effect can be achieved by configuring the model creation unit using other means, such as a neural network, to create the machine learning model. [Explanation of symbols]

[0086] 10 microphone, 20 computer, 21 spectrum analysis unit, 22 flame detection unit, 23 feature calculation unit, 24 support vector machine (model creation unit).

Claims

1. A microphone that collects sounds generated in the monitoring area as acoustic data, a level transition calculation unit that calculates a time transition of a sound pressure level in a predetermined high frequency band defined as a range of 3 kHz to 10 kHz from the acoustic data collected by the microphone; a flame detection unit that counts the number of times that the sound pressure level changes from a value below a predetermined judgment level to a value equal to or greater than the judgment level over time, and determines that a flame has occurred in the monitoring area due to a specific combustion material that generates a sudden sound when burning, if the number of times is equal to or greater than the predetermined judgment number; A fire detection device equipped with:

2. A microphone that collects sounds generated in the monitoring area as acoustic data, a level transition calculation unit that calculates a time transition of a sound pressure level in a predetermined high frequency band defined as a range of 3 kHz to 10 kHz from the acoustic data collected by the microphone; a feature calculation unit that counts the number of times the sound pressure level changes over time from a value less than a predetermined judgment level to a value equal to or greater than the judgment level, and outputs the counted number of times as a feature used to determine whether a flame has occurred due to a specific combustible material that generates a sudden sound when combusted; and a model creation unit that creates a machine learning model for determining whether a flame has been generated by a specific combustion material by using, as parameters, feature amounts calculated by the feature amount calculation unit for each of various acoustic data collected when a flame, which is a detection target, is generated in advance by burning a specific combustion material with a moisture content of 24% or more, which is more likely to generate sudden sounds during combustion, and various acoustic data collected when no flame has been generated, including false alarm factors; and a flame detection unit that uses the feature amount calculated by the feature amount calculation unit as an input to the machine learning model during monitoring of the monitoring area and determines whether a flame has occurred due to the specific combustion material; A fire detection device equipped with:

3. a sound collection step of collecting sounds generated in the monitoring area as acoustic data using a microphone; a level transition calculation step of calculating a time transition of a sound pressure level in a predetermined high frequency band defined as a range of 3 kHz to 10 kHz from the acoustic data collected by the microphone; a step of counting the number of times that the sound pressure level changes over time from a value below a predetermined judgment level to a value equal to or greater than the judgment level, and if the number of times is equal to or greater than the predetermined judgment number, identifying that a flame has occurred in the monitoring area due to a specific combustion material that generates a sudden sound when burning; A method for identifying a combustible material.

Images