CN111223265A

CN111223265A - Fire detection method, device, equipment and storage medium based on neural network

Info

Publication number: CN111223265A
Application number: CN202010297486.5A
Authority: CN
Inventors: 林永贤; 王昕�; 张�杰; 马启龙
Original assignee: Shanghai Aegis Industrial Safety Corp
Current assignee: Shanghai Aegis Industrial Safety Corp
Priority date: 2020-04-16
Filing date: 2020-04-16
Publication date: 2020-06-02
Anticipated expiration: 2040-04-16
Also published as: CN111223265B

Abstract

The invention discloses a fire detection method, a fire detection device, fire detection equipment and a fire detection storage medium based on a neural network, wherein the method comprises the following steps: acquiring incident light signals of a plurality of training light sources; performing signal processing and analysis on the incident light signal of each training light source to obtain a spectral signal parameter, a time signal parameter and a spatial signal parameter of each training light source; establishing a training data set according to the spectral signal parameter, the time signal parameter and the space signal parameter of the training light source, and establishing a neural network model for fire detection according to the training data set; and determining the output parameters of the light source of the field to be detected according to the neural network model, and determining whether the field to be detected has a fire or not according to the output parameters. According to the fire detection method provided by the embodiment of the invention, the neural network model is optimized through the characteristic parameters of the training light source, and the neural network model is utilized for fire identification, so that the identification algorithm is simple, the data processing amount is small, the response speed of fire detection is high, and the accuracy is high.

Description

Fire detection method, device, equipment and storage medium based on neural network

Technical Field

The embodiment of the invention relates to the technical field of fire detection, in particular to a fire detection method, a fire detection device, fire detection equipment and a fire detection storage medium based on a neural network.

Background

A fire refers to a disaster caused by combustion that is out of control in time or space. In the new standard, a fire is defined as a combustion that is out of control in time or space. There are many techniques for recognizing a fire, and in order to ensure a response speed of a fire detection technique, it is an important research direction to apply a neural network to the field of fire detection.

At present, fire detection methods based on neural network models generally perform fire identification based on image features, and have the following problems:

the existing fire detection method based on the neural network only carries out flame identification based on image characteristics, and due to the existence of a large number of 'flame-like' interference sources in the environment, such as a fluttering red flag, a swaying red scarf, false flames of an artificial fireplace, even spots scattered on the ground after sunlight penetrates through a forest, and the like, false alarm is easy to occur; the existing fire detection method based on the neural network has high requirements on software and hardware, a camera and video image analysis software need to be installed, the use cost is high, the detection area of the camera is small, the algorithm complexity is high, and the identification time is long; in order to improve the identification precision and the response speed, a large amount of test data is needed to establish a fire identification model, and the existing fire database, especially the database capable of clearly describing the characteristics of fire and flame, can hardly meet the requirement of training the model for a long time to achieve better detection performance.

Disclosure of Invention

The invention provides a fire detection method, which aims to solve the problems of complex fire identification algorithm, complex test data and high requirements on software and hardware, reduces the data required for building a fire identification model and improves the response speed of fire detection.

In a first aspect, an embodiment of the present invention provides a fire detection method based on a neural network, which specifically includes the following steps:

acquiring a plurality of training light sources;

acquiring an incident light signal of each training light source, performing signal processing and analysis on the incident light signal of each training light source, and acquiring a spectral signal parameter, a time signal parameter and a spatial signal parameter of each training light source, wherein the spectral signal parameter represents spectral intensity corresponding to optical signals with multiple preset wavelengths, the time signal parameter represents a trend of the spectral signal parameter corresponding to the optical signal with the same wavelength changing along with time, and the spatial signal parameter represents a trend of relative change of the spectral signal parameter corresponding to the optical signal with different wavelengths;

setting preset output parameters according to the training light source, and establishing a training data set according to the spectral signal parameters, the time signal parameters, the spatial signal parameters and the corresponding preset output parameters of the training light source;

establishing a neural network model, performing characteristic training on the neural network model according to the training data set, and determining the neural network model for fire detection;

and determining whether the fire occurs in the field to be detected according to the neural network model for fire detection.

In a second aspect, an embodiment of the present invention further provides a fire detection apparatus based on a neural network, including: the device comprises a signal processing unit and a control unit, wherein the signal processing unit comprises a filtering unit, an optical sensor and an analog-to-digital converter, the filtering unit is used for receiving an incident light signal and filtering the incident light signal to obtain optical signals with at least two preset wavelengths, the optical sensor is used for converting the optical signals into analog electric signals, and the analog-to-digital converter is used for converting the analog electric signals into digital signals; the control unit is used for analyzing an incident light signal of the training light source, and acquiring a spectral signal parameter, a time signal parameter and a spatial signal parameter of the training light source, wherein the spectral signal parameter represents spectral intensities corresponding to optical signals with multiple wavelengths, the time signal parameter represents a trend of the spectral signal parameter corresponding to the optical signal with the same wavelength changing along with time, the spatial signal parameter represents a trend of relative change of the spectral signal parameter corresponding to the optical signal with different wavelengths, a training data set is established according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and corresponding preset output parameters of the training light source, and a neural network model for fire detection is established according to the training data set; the control unit is also used for determining the calculation output parameters of the field to be detected according to the neural network model and determining whether the field to be detected has a fire or not according to the calculation output parameters.

In a third aspect, an embodiment of the present invention further provides an apparatus, including: one or more processors; a system memory to store one or more programs; when executed by the one or more processors, cause the one or more processors to implement the neural network-based fire detection method described above.

In a fourth aspect, the present invention further provides a storage medium, on which a computer program is stored, where the computer program is executed by a processor to implement the fire detection method based on a neural network.

The fire detection method based on the neural network of the embodiment of the invention constructs a training data set by characteristic parameters such as spectral signal parameters, time signal parameters, space signal parameters and the like of a training light source, trains a neural network model by using the training data set, carries out fire identification by using the output response of the neural network model to the characteristic parameters of the light source of a field to be detected, can rapidly distinguish flame forms and non-flame forms by combining the characteristic parameters such as the spectral signal parameters, the time signal parameters, the space signal parameters and the like, can establish the neural network model for fire detection by using a small amount of training data, has simple identification algorithm, small data processing amount, high response speed and high accuracy of fire detection, reduces the requirements on hardware performance, can realize the data analysis of the neural network model by using a smaller microprocessor, the practicability is strong.

Drawings

FIG. 1 is a flow chart of a fire detection method based on a neural network according to an embodiment of the present invention;

FIG. 2 is a flow chart of a fire detection method based on a neural network according to an embodiment of the present invention;

FIG. 3 is a flow chart of another fire detection method based on a neural network according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a neural network model for fire detection according to an embodiment of the present invention;

FIG. 5 is a flow chart of a method for fire detection based on a neural network according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a fire detection device based on a neural network according to a second embodiment of the present invention;

fig. 7 is a schematic structural diagram of an apparatus according to a third embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.

Because all fuels are burnt, the burning flame can generate form change and color change, and simultaneously emits heat outwards, the flame can be accurately identified by combining the jumping of the flame, the color of the flame and the heat of the flame, here, the spectrum signal, the time signal and the space signal of the flame are defined, the spectrum signal parameter, the time signal parameter and the space signal parameter of the flame form the characteristic parameter of the flame, and the specific meaning is as follows:

(1) spectral signature of flame

The spectrum refers to a pattern in which monochromatic light dispersed by a dispersion system (such as a prism and a grating) is sequentially arranged according to the wavelength (or frequency) after the monochromatic light is dispersed by the dispersion system, and is all called as an optical spectrum. The spectral range of the flame relates to near infrared rays, middle infrared rays, long infrared rays, ultraviolet rays and visible light, so that light emitted by the flame can be filtered to obtain optical signals with specific wavelengths (or frequencies), and the spectral intensity corresponding to the optical signals after filtering is used as the spectral signal parameters of the flame.

(2) Time signal of flame

The fuel burns in the air to generate flame, the flicker frequency of the flame is 3Hz to 8Hz, the oxygen concentration in the air changes along with the burning time when a fire breaks out, and the oxygen concentration in the air is lower when the time for the fire breaks out is longer. The color of the flame is changed under the influence of the oxygen concentration in the air, and under the condition of oxygen enrichment, the color of the flame is close to blue; the wavelength emitted by the flame changes along with the time, and the color of the flame is close to red under the anoxic state; therefore, the time signal parameter of the flame can be obtained according to the trend that the spectral intensity corresponding to the optical signal with any preset wavelength in the flame changes along with the time.

(3) Spatial signature of flame

When the fire is increased, the heat intensity of the flame is increased, the intensity of the corresponding spectrum signal is also increased, the heat intensity of the light source and the intensity of the spectrum signal generated by the interference sources such as lamplight and the like are not changed or only slightly changed, and therefore the trend that the spectrum signal parameters corresponding to the light signals with different preset wavelengths in the flame are relatively changed can be used as the space signal parameters of the flame.

The spectral signal parameter, the time signal parameter and the space signal parameter corresponding to the light source in the flame form and the non-flame form are obviously different, and the characteristic parameters such as the spectral signal parameter, the time signal parameter and the space signal parameter are combined with one another, so that the flame form and the non-flame form can be rapidly distinguished.

Based on this, the embodiments of the present invention provide a fire detection method, a fire detection apparatus, a device and a storage medium based on a neural network, and the fire detection method, the fire detection apparatus, the device and the storage medium based on the neural network according to the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

Example one

The fire detection method provided by the embodiment is suitable for application scenes for flame identification by utilizing a neural network model. In this embodiment, characteristic parameters of the training light source, such as a spectral signal parameter, a temporal signal parameter, and a spatial signal parameter, are used as input parameters of the neural network model, preset output parameters are set according to the type of the training light source, and the neural network model is trained by using the input parameters and the preset output parameters to construct the neural network model for fire detection. Fig. 1 is a flowchart of a fire detection method based on a neural network according to an embodiment of the present invention, which may be performed by a fire detection device storing a neural network model. As shown in fig. 1, the fire detection method based on the neural network specifically includes the following steps:

step S101: a plurality of training light sources is acquired.

When fire identification is performed, light sources in the surrounding environment, such as street lamps, car lamps, strong light sources, heat sources and the like, can interfere with identification of the fire, and the training light source is a typical light source which is manually set and has specific characteristic parameters, and typical flames and an ambient light source can be selected as the training light source.

Optionally, the training light source comprises a flame, an interference light source and a flame disposed under the interference light source, typically the interference light source comprises an arc light source, a heater light source and a lamp light source.

Wherein the flame may be a flame generated by combustion of any type of fuel, and the interfering light source may be a common light source that interferes with a characteristic parameter of the flame, such as one or more of an arc welding light source, a heater light source, and a halogen light source, wherein both the arc welding and the flame may emit ultraviolet light at 160 nm to 220 nm, the arc welding ultraviolet light emission being distinguished from the flame ultraviolet light emission in that the duration of the arc welding ultraviolet light emission generally does not exceed 1 second, and the flame ultraviolet light is emitted continuously during a fire; the heater and the halogen lamp and the flame both emit heat and light signals with different wavelengths, and the heater and the halogen lamp are different from the flame in that the wavelength and the heat intensity emitted by the heater and the halogen lamp do not change with time, the wavelength emitted by the flame changes with time, and the heat intensity of the flame also becomes larger with the increase of the fire. Therefore, the neural network model for fire detection can be constructed by learning and training the spectral signal parameters, the time signal parameters and the space signal parameters of different types of training light sources by means of the neural network model, the more the types of the training light sources are, the larger the corresponding data quantity is, the faster the response speed of fire identification is, and the higher the accuracy rate is.

Step S102: adjusting the distance between the flame in the training light source and the flame detection device, acquiring an incident light signal of each training light source, analyzing the incident light signal of each training light source, and acquiring a spectral signal parameter, a time signal parameter and a spatial signal parameter of each training light source.

The training light source under each distance parameter corresponds to one incident light signal, the spectral range of the incident light signal of each training light source relates to near infrared rays, middle infrared rays, long infrared rays, ultraviolet rays and visible light, filtering, photoelectric conversion and analog-to-digital conversion processing can be carried out on the incident light signal of the training light source, digital signals corresponding to the optical signals with different wavelengths (or frequencies) are obtained, and the spectral intensity corresponding to the digital signals corresponding to the optical signals with different wavelengths (or frequencies) is used as the spectral signal parameter of the training light source. And judging whether the numerical value of the spectral signal parameter corresponding to the optical signal with the same wavelength in the training light source changes along with the time, and taking the increment of the spectral signal parameter of the training light source changing along with the time as the time signal parameter of the training light source. And judging whether the relative values among the spectral signal parameters corresponding to the optical signals with different wavelengths in the training light source change or not, and taking the relative values among the different spectral signal parameters representing the training light source as the space signal parameters of the flame.

In this embodiment, the spectral intensity, color and heat of the flame are changed in a large range and the spectral intensity, color and heat of the interfering light source are changed in a small or even unchanged range over time, i.e., if the training light source comprises a flame, the change rate of the time signal parameter is greater than the preset change threshold, the value of the space signal parameter is greater than the preset ratio threshold, therefore, by learning the spectral signal parameter, the time signal parameter and the space signal parameter of the training light source, the neural network model can obviously distinguish the fire form from the non-fire form, reduce the training data of the neural network model, the incident light signal can be obtained by simple components such as a filter, a light sensor, an analog-to-digital converter and the like, and incident light signals are analyzed by using a simple algorithm to obtain the parameters, so that the difficulty of executing the algorithm is reduced.

Step S103: and establishing a training data set according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and the corresponding preset output parameter of the training light source.

The preset output parameters can be set according to the types of the training light sources, typically, the action weight of flame on the neural network model can be set to be higher than the action weight of the interference light sources on the neural network model, the action weights of different types of interference light sources on the neural network model are set to be the same, and when no training light source is set, the preset output parameters are set to be 0. For example, if the training light source is a single interference light source, the preset output parameters corresponding to the interference light sources may be set to be preset interference values, for example, the preset interference value may be 0.5; if the training light source is an independent flame, the preset output parameter can be set to a preset flame minimum value, for example, the preset flame minimum value can be 2; if the training light source is flames under the environment of N interference sources, the preset output parameter can be set to be N × preset interference value + preset flame minimum value.

In this embodiment, the spectral signal parameter, the temporal signal parameter, and the spatial signal parameter of the training light source are used as input parameters of the neural network model, each set of input parameters and corresponding output parameters form a set of training data, a plurality of training data can be obtained by setting different types of training light sources, adjusting the distance between the training light source and the flame detection device, and the plurality of training data form a training data set.

Step S104: a neural network model for fire detection is constructed from the training dataset.

In this embodiment, the neural network model for fire detection may be constructed by performing optimization iteration on the general neural network model according to the training data set. The neural network model is trained through a training data set established based on the spectral signal parameters, the time signal parameters and the space signal parameters, the neural network model for fire detection can obviously distinguish fire states from non-fire states, and the accuracy of fire identification can be improved.

Step S105: and determining a calculation output parameter of a light source of the field to be detected according to the neural network model for fire detection, and determining whether the field to be detected has a fire or not according to the calculation output parameter.

The site to be detected is a site where fire detection is required, such as a residence, a commercial district, a warehouse, and the like. The method can acquire characteristic parameters of the light source of the field to be detected, such as spectral signal parameters, time signal parameters and space signal parameters, and uses the characteristic parameters of the light source of the field to be detected as input parameters of the neural network model, calculates the output parameters of the light source of the field to be detected by using the neural network model for fire detection, and judges whether the field to be detected is subjected to fire identification according to specific values of the output parameters.

Therefore, according to the fire detection method provided by the embodiment of the invention, the neural network model is optimized through the characteristic parameters of the training light source, the characteristic parameters of the light source of the field to be detected are input into the neural network model for calculation, and fire identification is carried out according to the calculated output parameters.

Fig. 2 is a flowchart of a fire detection method based on a neural network according to an embodiment of the present invention. Fig. 2 shows a specific process of acquiring a spectral signal parameter, a temporal signal parameter, and a spatial signal parameter corresponding to each training light source in the first embodiment of the present invention.

Optionally, as shown in fig. 2, the obtaining an incident light signal of each training light source, analyzing the incident light signal of each training light source, and obtaining a spectral signal parameter, a temporal signal parameter, and a spatial signal parameter of each training light source includes the following steps:

step S201: and acquiring at least one light signal with preset frequency emitted by each training light source.

In this embodiment, the obvious characteristic that the flame is distinguished from the interference light source is that the light signal emitted by the flame flickers at a frequency of 3hz to 8hz, the fluctuation of the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and the like corresponding to the flame is large, and the fluctuation of the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and the like corresponding to the interference light source in the non-flame form is large, so that the preset frequency can be set to be 5hz, for example, the light signal with the preset frequency emitted by the training light source is acquired, the variation trend of the light signal with the preset frequency is studied, and the flame form and the non-flame form can be distinguished obviously.

Certainly, in a windy state, the flame jumping frequency can be increased, so that different preset frequencies can be set, the change trend of the optical signal under each preset frequency is researched and compared, the data volume of the input parameters of the neural network model is enlarged, and the accuracy of the neural network model identification is improved.

Step S202: and filtering the incident light signal of each preset frequency to obtain light signals of at least two preset wavelengths.

For example, the filtering process may be performed on the incident light signal of the preset frequency by arranging different light sensors, which allow the incident light signal of a specific wavelength to pass through, and for facilitating the data analysis, arranging at least two light sensors of preset wavelengths (for example, a first preset wavelength and a second preset wavelength) to perform the filtering process on the incident light signal.

Illustratively, the main components decomposed during fuel combustion are carbon and hydrogen, carbon burns in oxygen to form carbon dioxide, and emits a spectral signal with a wavelength of 4300 nm; hydrogen burns in oxygen to form water molecules and emits a spectrum signal with the wavelength of 2700 nanometers; in addition, since the light source has a specific heat, the light emitted by the flame combustion does not include a spectrum of 3800 nm, and a spectrum signal having a wavelength of 3800 nm can be used as a heat reference. A first preset wavelength (for example, 4300 nm) optical sensor and a second preset wavelength (for example, 3800 nm) optical sensor may be configured to perform filtering processing on an incident light signal, so as to obtain an optical signal with a wavelength of 4300 nm and an optical signal with a wavelength of 3800 nm; alternatively, a preset first preset wavelength (for example, 4300 nm) light sensor, a second preset wavelength (for example, 3800 nm) light sensor and a third preset wavelength (for example, 2700 nm) light sensor may be arranged to perform filtering processing on the incident light signal, so as to obtain an optical signal with a wavelength of 4300 nm, an optical signal with a wavelength of 2700 nm and an optical signal with a wavelength of 3800 nm.

Step S203: and carrying out photoelectric conversion and analog-to-digital conversion processing on the optical signal with each preset wavelength to obtain the spectral signal parameters of the training light source.

The optical signal can be converted into an analog electrical signal by the optical sensor, and the analog electrical signal is converted into a digital signal by the analog-to-digital converter, so that the digital signal corresponding to the optical signal with each preset wavelength can be obtained, and the digital signal corresponding to the optical signal with any preset frequency and preset wavelength in the training light source is used as a spectral signal parameter.

Step S204: and respectively carrying out derivation operation on each spectral signal parameter to obtain a time signal parameter of the training light source.

In this embodiment, a time derivative of the spectral signal parameter of each training light source is calculated, where the time derivative of the spectral signal parameter of the training light source is an increment of the spectral signal of the training light source along with the time change. If the training light source comprises a flame configuration, the time derivative of the spectral signal parameter of the training light source is greater than the time derivative of the spectral signal parameter of the training light source in a non-flame configuration.

Step S205: and calculating the ratio of the spectral signal parameters to obtain the spatial signal parameters of the training light source.

For example, the optical signal with the first preset frequency (for example, the first preset frequency is 5 hz) may be filtered to obtain optical signals with two preset wavelengths (for example, the first preset wavelength and the second preset wavelength), each of the optical signals with the preset wavelengths corresponds to one of the spectral signal parameters, for example, the optical signals with the first preset frequency and the first preset wavelength correspond to the first spectral signal parameter X1, the optical signals with the first preset frequency and the second preset wavelength correspond to the second spectral signal parameter X2, the ratio of the first spectral signal parameter X1 to the second spectral signal parameter X2 is calculated, and the change of the ratio of the first spectral signal parameter X1 to the second spectral signal parameter X2 is determined. If the training light source is a flame, the rate of change of the ratio is large.

The following describes the process of resolving an incident optical signal with reference to a specific example.

For example, three optical sensors may be disposed on the incident light signal receiving side of the fire detection device to process incident light signals of two predetermined frequencies (e.g., a first predetermined frequency 5HZ and a second predetermined frequency, e.g., 25 HZ), and each optical sensor may have a surface with different spectral filters to obtain light signals of three predetermined wavelengths, for example, a first spectral filter is used to filter a spectral signal with a wavelength of 4300 nm, a second spectral filter is used to filter a spectral signal with a wavelength of 3800 nm, and a third spectral filter is used to filter a spectral signal with a wavelength of 2700 nm, where the applicable frequency of the spectral filters is 5HZ to 25HZ, so as to prevent low-frequency noise from entering the fire detection device and improve the accuracy of fire detection.

(1) Obtaining spectral signal parameters of a training light source

In this embodiment, a spectral filter is adopted to obtain optical signals of a first preset frequency and a first preset wavelength, optical signals of the first preset frequency and a second preset wavelength, and optical signals of the first preset frequency and a third preset wavelength, an optical sensor is adopted to perform photoelectric conversion on the optical signals output by the spectral filter, converted analog electrical signals are sent to an analog-to-digital converter, the analog-to-digital converter is adopted to convert the analog electrical signals into digital signals, and the optical signals of the first preset frequency and the first preset wavelength are subjected to photoelectric conversion and analog-to-digital conversion processing to obtain first spectral signal parameters; performing photoelectric conversion and analog-to-digital conversion on the optical signals with the first preset frequency and the second preset wavelength to obtain a second spectrum signal parameter; performing photoelectric conversion and analog-to-digital conversion on the optical signals with the first preset frequency and the third preset wavelength to obtain a third spectral signal parameter; performing photoelectric conversion and analog-to-digital conversion on the optical signals with the second preset frequency and the first preset wavelength to obtain fourth spectral signal parameters; performing photoelectric conversion and analog-to-digital conversion on the optical signals with the second preset frequency and the second preset wavelength to obtain fifth spectral signal parameters; and performing photoelectric conversion and analog-to-digital conversion on the optical signals with the second preset frequency and the third preset wavelength to obtain a sixth spectral signal parameter.

Exemplarily, a first digital signal corresponding to a 5hz 4300 nm optical signal is taken as the first spectral signal parameter X1; taking a second digital signal corresponding to the optical signal with the wavelength of 5Hz and the wavelength of 3800 nm as a second spectral signal parameter X2; taking a third digital signal corresponding to the optical signal with the wavelength of 5Hz and the wavelength of 2700 nm as a third spectral signal parameter X3; taking a fourth digital signal corresponding to the 25Hz 4300 nm spectral signal as a fourth spectral signal parameter X4; taking a fifth digital signal corresponding to the 25Hz 3800 nm spectral signal as a fifth spectral signal parameter X5; the sixth digital signal corresponding to the 25hz 2700 nm spectral signal is taken as the sixth spectral signal parameter X6.

The first to sixth spectral signal parameters X1 to X6 are used to characterize the spectral intensities of the optical signals with different frequencies and different wavelengths.

(2) Obtaining time signal parameters of a training light source

In this embodiment, a first spectral signal parameter X1 corresponding to an optical signal with a first preset frequency and a first preset wavelength, a second spectral signal parameter X2 corresponding to an optical signal with a first preset frequency and a second preset wavelength, and a third spectral signal parameter X3 corresponding to an optical signal with a first preset frequency and a third preset wavelength are respectively obtained, where the first preset wavelength, the second preset wavelength, and the third preset wavelength may be adjusted according to a fuel type of a site to be measured; calculating the time derivative of the first spectral signal parameter X1

Acquiring a first time signal parameter X7, wherein the first time signal parameter X7 is used for representing the increment of the change of the first spectral signal parameter X1 along with time; calculating the time derivative of the second spectral signal parameter X2

Acquiring a second time signal parameter X8, wherein the second time signal parameter X8 is used for representing the increment of the change of the second spectral signal parameter X2 along with time; calculating the time derivative of the third spectral signal parameter X3

A third time signal parameter X9 is obtained, the third time signal parameter X9 being used to represent the delta in the change over time of the third spectral signal parameter X3.

The first spectral signal parameter X1 and the third spectral signal parameter X3 may be obtained from an optical signal emitted by a flame-shaped training light source, the second spectral signal parameter X2 may be obtained from an optical signal emitted by a non-flame-shaped training light source, the first spectral signal parameter X1 to the third spectral signal parameter X3 correspond to an optical signal with a first preset frequency, and the first preset frequency may be a normal frequency of a flame, for example, the first preset frequency may be 5 hz. The first time signal parameter X7 is used for representing the increment of the spectral intensity of the optical signal with the first preset wavelength in the optical signal with the first preset frequency along with the change of time; the second time signal parameter X8 is used for characterizing the increment of the spectral intensity of the optical signal with the second preset wavelength in the optical signal with the first preset frequency along with the change of the time; the third time signal parameter X9 is used to represent an increment of the spectral intensity of the optical signal with the third preset wavelength in the optical signal with the first preset frequency along with the change of time, and the first preset wavelength, the second preset wavelength, and the third preset wavelength may be adjusted according to the fuel type of the site to be measured. Over time, the spectral intensity and the color change amplitude of the flame are large, the spectral intensity change amplitude of the interference light source is small, even the spectral intensity change amplitude of the interference light source does not change, therefore, the time-varying increment of at least one of the first time signal parameter X7 to the third time signal parameter X9 corresponding to the training light source in the flame form is larger than the preset increment threshold, the first time signal parameter X7 to the third time signal parameter X9 corresponding to the training light source in the non-flame form are approximately 0, and the first time signal parameter X7 to the third time signal parameter X9 are used as the input parameters of the neural network model, so that the difference of the calculation output parameters corresponding to different types of training light sources can be increased, and the accuracy of identifying the flame form is improved.

Of course, the time derivatives of the first spectral signal parameter X1 to the sixth spectral signal parameter X6 may also be calculated according to the requirement, and the calculation result is added to the training data set as the input parameter of the neural network model, which is not limited to this, and the more the input parameters in the training data set, the higher the detection accuracy of the constructed neural network model for fire detection, the larger the workload, and the number of the input parameters may be adjusted according to the cost and the effect.

(3) Obtaining spatial signal parameters of a training light source

In this embodiment, a first spectral signal parameter X1 corresponding to an optical signal with a first preset frequency and a first preset wavelength, a second spectral signal parameter X2 corresponding to an optical signal with a first preset frequency and a second preset wavelength, and a third spectral signal parameter X3 corresponding to an optical signal with a first preset frequency and a third preset wavelength are respectively obtained, where the first preset wavelength, the second preset wavelength, and the third preset wavelength may be adjusted according to a fuel type of a site to be measured; calculating the ratio of the first spectral signal parameter X1 to the second spectral signal parameter X2

Acquiring a first spatial signal parameter X10; calculating the ratio of the first spectral signal parameter X1 to the third spectral signal parameter X3

Acquiring a second spatial signal parameter X11; calculating the average value of the ratio of the first spectral signal parameter X1, the third spectral signal parameter X3 and the second spectral signal parameter X2

And a third spatial signal parameter X12 is obtained.

The first spectral signal parameter X1 and the third spectral signal parameter X3 may be obtained from light signals emitted by a flame-shaped training light source, and the second spectral signal parameter X2 may be obtained from light signals emitted by a non-flame-shaped training light source. The values of the first space signal parameter X10 and the third space signal parameter X12 corresponding to the training light source in the flame form are larger than the values of the first space signal parameter X10 and the third space signal parameter X12 corresponding to the training light source in the non-flame form; the first spectral signal parameter X1 and the third spectral signal parameter X3 have different values in flames produced by the combustion of different types of fuels, for example, the first spectral signal parameter X1 has a value larger than that of the third spectral signal parameter X3 in flames produced by using gasoline as a fuel; the value of the first spectral signal parameter X1 in the hydrogen fire is smaller than the value of the third spectral signal parameter X3, and the first spatial signal parameter X10 to the third spatial signal parameter X12 are used as input parameters of a neural network model, so that the difference of the calculation output parameters corresponding to different types of training light sources can be increased, and the accuracy of flame form identification is improved.

Further, the first to sixth spectral signal parameters X1 to X6, the first to third temporal signal parameters X7 to X9, and the first to third spatial signal parameters X10 to X12 may be input into the neural network model as input parameters, each set of input parameters is correspondingly set with a preset output parameter, and the preset output parameters are adjusted and set according to the type of the training light source.

It should be understood that since the spectral signal intensity and the heat intensity of the flame itself are greater than those of the interference light source, the value of the first spatial signal parameter X10 with a flame in the training light source is greater than the value of the first spatial signal parameter X10 without a flame in the training light source.

Of course, two light sensors may be provided on the incident light signal receiving side of the fire detection device to process incident light signals of two predetermined frequencies (e.g., a first predetermined frequency of 5HZ and a second predetermined frequency of 25 HZ).

Specifically, the surface of each photosensor is provided with different spectral filters, for example, a first spectral filter for filtering spectral signals having a wavelength of 4300 nm and a second spectral filter for filtering spectral signals having a wavelength of 3800 nm.

(1) Obtaining spectral signal parameters of a training light source

In this embodiment, the optical sensor performs photoelectric conversion on a spectral signal output by the spectral filter, sends the converted analog electrical signals to the analog-to-digital converter, converts each analog electrical signal into a digital signal by using the analog-to-digital converter, and takes a first digital signal corresponding to a 5hz 4300 nm spectral signal as a first spectral signal parameter X1; taking a second digital signal corresponding to the 5Hz 3800 nm spectral signal as a second spectral signal parameter X2; taking a third digital signal corresponding to the 25Hz 4300 nm spectral signal as a third spectral signal parameter X3; the fourth digital signal corresponding to the 25hz 3800 nm spectral signal is taken as the fourth spectral signal parameter X4.

(2) Obtaining time signal parameters of a training light source

In the present embodiment, the time derivative of the first spectral signal parameter X1 is calculated

Acquiring a first time signal parameter X5, wherein the first time signal parameter X5 is used for representing the increment of the change of the first spectral signal parameter X1 along with time; calculating the time derivative of the second spectral signal parameter X2

A second time signal parameter X6 is obtained, the second time signal parameter X6 being indicative of an increment 6 in the change over time of the second spectral signal parameter X2.

(3) Obtaining spatial signal parameters of a training light source

In the present embodiment, the ratio of the first spectral signal parameter X1 to the second spectral signal parameter X2 is calculated

A first spatial signal parameter X7 is obtained.

Further, the first to fourth spectral signal parameters X1 to X4, the first temporal signal parameter X5, the second temporal signal parameter X6 and the first spatial signal parameter X7 may be input into the neural network model as input parameters, each set of input parameters corresponds to an output parameter, and the output parameters may be adjusted and set according to the type of the training light source.

Fig. 3 is a flowchart of another fire detection method based on a neural network according to an embodiment of the present invention. Fig. 4 is a schematic structural diagram of a neural network model for fire detection according to an embodiment of the present invention.

Optionally, as shown in fig. 3, the establishing a neural network model, performing feature training on the neural network model according to the training data set, and determining the neural network model for fire detection includes the following steps:

step S301: and establishing a neural network model, wherein the neural network model comprises a weight matrix and a deviation matrix.

Illustratively, as shown in fig. 4, a neural network model with 2 layers of neural nodes may be selected, a first neural layer sets 10 nodes, a second neural layer sets 5 nodes, the neural network model includes each layer of weight matrix (a first weight matrix W1, a second weight matrix W2, and a third weight matrix W3) and each layer of bias matrix (a first bias matrix B1 and a second bias matrix B2), and optimization of each layer of weight matrix and each layer of bias matrix may be implemented.

Step S302: and performing iterative training on the neural network model according to the training data set, constructing a loss function, detecting the value of the loss function once every iteration, and if the value of the loss function is smaller than a preset convergence threshold value, acquiring a converged weight matrix and a converged deviation matrix.

Exemplarily, as shown by referring to fig. 4, the first to sixth spectral signal parameters X1 to X6, the first to third temporal signal parameters X7 to X9, and the first to third spatial signal parameters X10 to X12 corresponding to any training light source are obtained, the first to sixth spectral signal parameters X1 to X6, the first to third temporal signal parameters X7 to X9, the first to third spatial signal parameters X10 to X12, and the like are input into the neural network model as the input parameters X, and the calculation output parameters of the neural network model are calculated.

Because the input parameters comprise twelve parameters, and the first neural layer is provided with 10 nodes, the first weight matrix W1 is constructed into a data matrix with ten rows and twelve columns, and the product of the first weight matrix W1 and the input parameters X is calculated to obtain the output matrix L1 of the neurons in the first layer, wherein the functional expression of the output matrix L1 is shown in formula I.

(formula one)

Further, the sum of the output matrix L1 of the first layer neurons and the first bias matrix B1 is calculated and all inputs are compressed below the limit value, resulting in the input matrix Q1 of the first layer neurons, whose functional expression is shown in equation two.

(formula two)

Further, since the first neural layer is provided with 10 nodes, and the second neural layer is provided with 5 nodes, the second weight matrix W2 is constructed as a data matrix with five rows and ten columns, and the product of the second weight matrix W2 and the input matrix Q1 of the first layer neuron is calculated to obtain the output matrix L2 of the second layer neuron, wherein the functional expression of the output matrix L2 is shown in formula three.

(formula three)

Further, the sum of the output matrix L2 of the second layer neurons and the second deviation matrix B2 is calculated, and all inputs are compressed below the limit value, resulting in the output matrix Q2 of the second layer neurons, whose functional expression is shown in equation four.

(formula four)

Further, calculating the product of the third weight matrix W3 and the output matrix Q2 of the second layer neuron to obtain the calculated output parameter Z of the neural network model, wherein the functional expression of the calculated output parameter Z is shown in formula five.

(formula five)

In this embodiment, the purpose of optimizing the neural network model is to reduce the difference between the preset output parameter Y and the calculated output parameter Z, sum the difference between the preset output parameter and the calculated output parameter in all training data sets, establish a loss function of the neural network model, iteratively update each layer of weight matrix and each layer of bias matrix by using a gradient descent algorithm, adjust the first weight matrix W1, the second weight matrix W2, the third weight matrix W3, the first bias matrix B1, and the second bias matrix B2, and if the value of the loss function is smaller than a preset convergence threshold, it indicates that the loss function converges, and the values of the first weight matrix W1, the second weight matrix W2, the third weight matrix W3, the first bias matrix B1, and the second bias matrix B2 when the loss function converges are weight matrices and bias matrices with excellent convergence.

Step S303: and constructing a neural network model for fire detection according to the converged weight matrix and the deviation matrix.

In this embodiment, a first weight matrix W1, a second weight matrix W2, a third weight matrix W3, a first deviation matrix B1, and a second deviation matrix B2 corresponding to the loss function convergence are substituted into the neural network model to establish an optimized neural network model, and the neural network model can be used for analyzing and calculating input data of a site to be detected, obtaining a calculation output parameter, and realizing fire detection by using the calculation output parameter.

In this embodiment, the training light source includes a flame, an interference light source, and a flame disposed under the interference light source, and the interference light source includes an arc welding light source, a heater light source, and a lamp light source. And constructing a neural network model for fire detection based on the training light source. Of course, the flame may be generated by burning different types of fuels, and the disturbing light source may be other light sources or energy sources disturbing fire detection, which is not limited to this, and the above training light source is merely used as an example to explain the working principle of the embodiment of the present invention.

Illustratively, the distance between the flame in the training light source and the flame detection device can be adjusted, a plurality of distance parameters are obtained, and a plurality of incident light signals corresponding to each type of training light source under any distance parameters are obtained. For example, the distance between the flame and the flame detection device in the training light source can be adjusted to obtain one incident light signal every 10 feet, for example, the distances between the flame and the flame detection device in the training light source are set to 10 feet, 20 feet, … …, 190 feet and 200 feet, and incident light signals corresponding to different distances are obtained respectively. Of course, the distance between the flame and the flame detection device in the training light source may be any data representing the distance, and is not limited thereto. If the training light source does not include flame, the distance parameter can be set to any value within a preset distance range, and in order to ensure the identity of the training data, the setting rule of the distance parameter can be kept consistent.

Setting distance parameters, acquiring corresponding incident light signals according to light signals emitted by any training light source under any distance parameter, analyzing each incident light signal respectively, acquiring input parameters (including spectral signal parameters, time signal parameters and space signal parameters) of multiple groups of flames and input parameters (including spectral signal parameters, time signal parameters and space signal parameters) of multiple groups of interference light sources, and a plurality of groups of input parameters (including spectral signal parameters, time signal parameters and space signal parameters) of flames under the interference light source, wherein the types of the training light source and the distance parameters between the training light source and the flame detection device are respectively used as design variables, the input parameters and preset output parameters are used as design target quantities, a training data set shown in table 1 is established, and a neural network model for fire detection is established according to the training data set.

TABLE 1

As can be seen from table 1, when the same training light source is disposed at different positions, a set of input parameters (e.g., X1-X12) is obtained for any training light source under any distance parameter, and if a preset number (e.g., 20) of distance parameters are set, the number of tests is a preset number (e.g., 20), and the number of input parameter sets is a preset number (e.g., 20). The preset output parameters are only related to the type of the training light source and are not related to the distance parameters. For example, if the training light source is a flame, the preset output parameters of the training light source corresponding to any distance parameter may be set to be a value a (e.g., a = 2); if the training light source is the interference light source 1, the preset output parameter corresponding to any distance parameter of the training light source can be set to be a value b (for example, b = 0.5); if the training light source is the interference light source 2, the preset output parameter corresponding to any distance parameter of the training light source can be set to be a value c (for example, c = 0.5); if the training light source is an interference light source 1+ an interference light source 2, the preset output parameter corresponding to any distance parameter of the training light source can be set to be a value b + c; if the training light source is the flame + interference light source 1, the preset output parameter corresponding to any distance parameter of the training light source can be set to be a value a + b; if the training light source is the flame + interference light source 2, the preset output parameter corresponding to any distance parameter of the training light source can be set as a value a + c; if the training light source is flame + interference light source 1+ interference light source 2, the preset output parameter corresponding to any distance parameter of the training light source can be set to be a value a + b + c.

For example, the values of the preset output parameters corresponding to different types of single interference light sources under different distance parameters may be set to be the same value, that is, the value b and the value c may be equal.

It should be noted that, the training of the neural network model can be completed through the 140 sets of training data, and the neural network model can accurately recognize the flame form in the interference environment. Of course, in order to improve the accuracy rate of flame identification, the size of the flame can be adjusted, and a plurality of groups of corresponding training data are set to supplement the training data set.

Typically, several types of training light sources can be provided: firstly, flame generated when n-heptane is used as fuel for combustion is used as a flame light source in a training light source, a preset output parameter of a neural network model to the flame is set as a, and a first training data set shown in a table 2 is established; secondly, an arc welding interference light source, a heater interference light source or a halogen lamp interference light source is used as an interference light source in the training light source, the preset output parameters of the neural network model to the single interference light source can be set to be the same (for example, b), and a second training data set shown in table 3 is established; and thirdly, establishing a third training data set shown in the table 4 by using a light source formed by flame generated by burning n-heptane under the environment of the interference light source, wherein the first training data set, the second training data set and the third training data set jointly form a training data set of the neural network model.

TABLE 2

As shown in table 2, the first column indicates the sequence numbers of each set of training data; the second column indicates that the training light source is a flame generated when burning with n-heptane as fuel; the third column shows the distance parameter between the fuel combustion point and the fire detection device, acquiring an incident light signal every 10 feet; the fourth column represents the number of training data sets obtained when any one of the training light sources sets any one of the distance parameters; the fifth column represents a set of input parameters (e.g., X1-X12) corresponding to any one of the training light sources at any one of the distance parameters; the sixth column represents preset output parameters corresponding to any training light source under any distance parameter, a group of input parameters in each row and the corresponding preset output parameters form a training data set, the group of input parameters are substituted into the neural network model, the neural network model outputs calculation output parameters after operation processing, and the neural network model is trained by adjusting the difference value of the preset output parameters and the calculation output parameters.

TABLE 3

As shown in table 3, the first column indicates the sequence numbers of each set of training data; the second column indicates that the training light source is a different interference light source; the third column indicates a randomly set distance parameter; the fourth column represents the number of training data sets obtained when any one of the training light sources sets any one of the distance parameters; the fifth column represents a set of input parameters (e.g., X1-X12) corresponding to any one of the training light sources at any one of the distance parameters; the sixth column represents preset output parameters corresponding to any training light source under any distance parameter, a group of input parameters in each row and the corresponding preset output parameters form a training data set, the group of input parameters are substituted into the neural network model, the neural network model outputs calculation output parameters after operation processing, and the neural network model is trained by adjusting the difference value of the preset output parameters and the calculation output parameters. Because the physical effects of the heater and the halogen lamp are similar, the halogen lamp in table 3 can be replaced by the heater, and the test result is not changed.

TABLE 4

As shown in table 4, the first column indicates the sequence numbers of each set of training data; the second column represents the flame generated by burning n-heptane under different interference light sources; the third column shows the distance parameter between the fuel combustion point and the fire detection device, acquiring an incident light signal every 10 feet; the fourth column represents the number of training data sets obtained when any one of the training light sources sets any one of the distance parameters; the fifth column represents a set of input parameters (e.g., X1-X12) corresponding to any one of the training light sources at any one of the distance parameters; the sixth column represents preset output parameters corresponding to any training light source under any distance parameter, a group of input parameters in each row and the corresponding preset output parameters form a training data set, the group of input parameters are substituted into the neural network model, the neural network model outputs calculation output parameters after operation processing, and the neural network model is trained by adjusting the difference value of the preset output parameters and the calculation output parameters. Because the physical effects of the heater and the halogen lamp are similar, the halogen lamp in table 4 can be replaced by the heater, and the test result is not changed.

As shown in tables 1 to 4, the fuel for generating the flame in the training light source can be any type of fuel, the main components of n-heptane (C7H 16) include carbon and hydrogen, the flame generated by the combustion of the fuel is the same as the flame generated by the combustion of most of the fuels (such as wood, cloth, paper and the like), and the flame generated by using n-heptane as the base fuel can replace the flame generated by most of the fuels, so that the data volume in the training data set is reduced, and the model optimization time is shortened.

Of course, other types of fuels can be added to train the neural network model based on the training data set of table 1.

In this embodiment, after constructing the neural network model for fire detection from the training data set, the following steps are further included: taking flame generated by burning fuel in a field to be tested in an interference-free light source environment as a supplementary training light source, wherein the fuel type of the supplementary training light source is different from that of the training light source; adjusting the distance between the supplementary training light source and the flame detection device, acquiring incident light signals of a plurality of supplementary training light sources, and performing signal processing and analysis on the incident light signals of each supplementary training light source to acquire spectral signal parameters, time signal parameters and space signal parameters of the supplementary training light sources; establishing a supplementary training data set according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and the corresponding supplementary preset output parameter of the supplementary training light source; and optimizing the neural network model according to the supplementary training data set and the training data set.

For example, the fuel of the site to be tested can be gasoline, flame generated by gasoline combustion is used as a supplementary training light source in an environment without interference with a light source, the distance between the gasoline combustion point and the flame detection device is adjusted, for example, the distance parameter between the gasoline combustion point and the fire detection device is set every 10 feet, and 20 groups of distance parameters are set, for example, 10 feet, 20 feet, … … and 200 feet. Acquiring incident light signals of 20 groups of gasoline flames, analyzing the incident light signals, acquiring spectral signal parameters, time signal parameters and spatial signal parameters of the 20 groups of gasoline flames, substituting the spectral signal parameters, the time signal parameters and the spatial signal parameters of the gasoline flames into a neural network model, setting a complementary preset output parameter Y ' corresponding to the gasoline flames, and constructing a complementary training data set of the gasoline flames under different distance parameters, wherein the complementary training data set comprises 20 groups of training data 1# ' { X1-X12, Y ' }, 2# ' { X1-X12, Y ' }, … …, and 20# ' { X1-X12, Y ' }.

Further, for flames generated by different types of fuels, under the same distance parameter, a first spectral signal parameter corresponding to the flame a in an interference-free light source environment has the same proportional relationship with the sum of a preset interference light source B and a first spectral signal parameter corresponding to the flame a of the same preset interference light source B. And establishing a preset factor K based on the proportional relation, wherein the supplemented preset output parameter Y 'corresponding to flame generated by gasoline combustion in the environment with a preset interference light source B (such as electric arc welding) can be estimated by multiplying the supplemented preset output parameter Y' by the preset factor K.

For example, taking the distance parameter of 200 feet as an example, in conjunction with the data in tables 2-4, the predetermined factor K may be expressed by equation six as shown below,

(formula six)

Wherein the content of the first and second substances,

for the first spectral signal parameter in the training data set numbered 20 in table 2,

indicating a clean light source environment with a distance parameter of 200 feetThe corresponding first spectral signal parameter is then compared to the first spectral signal parameter,

for the first spectral signal parameter in the training data set numbered 100 in table 4,

a first spectral signal parameter indicative of an n-heptane flame in an environment of an arc welding interference light source at a distance parameter of 200 feet,

for the first spectral signal parameter in the training data set numbered 40 in table 3,

a first spectral signal parameter indicative of a corresponding arc welding interference light source at a distance parameter of 200 feet.

Therefore, by introducing the preset factor K, the corresponding supplement preset output parameters can be obtained through calculation, the neural network model constructs a loss function by utilizing the supplement preset output parameters and the calculation output parameters, iterative updating is carried out until the loss function is converged, training of the neural network model for fire detection can be completed without establishing training data of gasoline flame in an interference environment to train the neural network model, and the quantity of artificially obtained training data can be reduced.

It should be noted that the more data in the training data table, the higher the accuracy of the neural network model for fire detection is, and the lower the false alarm rate for fire detection is.

Fig. 5 is a flowchart of another fire detection method based on a neural network according to an embodiment of the present invention.

Optionally, as shown in fig. 5, determining an output parameter of a light source of the field to be tested according to the neural network model, and determining whether a fire occurs in the field to be tested according to the output parameter, includes the following steps:

step S401: and acquiring an incident light signal of a light source of the field to be detected.

The site to be detected is a site needing fire detection, such as a residence, a commercial district, a storehouse and the like, and a light source of the site to be detected is analyzed to judge whether flames exist in the light source of the site to be detected.

Step S402: analyzing an incident light signal of a light source of the field to be detected, and acquiring a spectrum signal parameter, a time signal parameter and a space signal parameter of the light source of the field to be detected.

In this embodiment, an incident light signal of a light source in a field to be measured is analyzed to obtain a spectrum signal parameter, a time signal parameter, and a spatial signal parameter of the light source in the field to be measured, and the method specifically includes the following steps: filtering incident light signals of a light source of a field to be detected to obtain optical signals with at least two preset wavelengths; performing photoelectric conversion and analog-to-digital conversion processing on each optical signal with preset wavelength to obtain spectral signal parameters of a light source of a field to be detected; acquiring a plurality of spectral signal parameters, and respectively carrying out derivation operation on each spectral signal parameter to acquire a time signal parameter of a light source of a field to be detected; spectrum signal parameters corresponding to a plurality of optical signals with different preset wavelengths are obtained, the ratio of each spectrum signal parameter is calculated, and the space signal parameters of the light source of the field to be measured are obtained.

For example, three optical sensors may be disposed on the incident light signal receiving side of the fire detection device to process incident light signals of two preset frequencies (e.g., a first preset frequency of 5HZ and a second preset frequency of 25HZ, for example), and each optical sensor may have a surface provided with a different spectral filter, for example, a first spectral filter for filtering a spectral signal with a wavelength of 4300 nm, a second spectral filter for filtering a spectral signal with a wavelength of 3800 nm, and a third spectral filter for filtering a spectral signal with a wavelength of 2700 nm, where the applicable frequency of the spectral filter is 5HZ to 25HZ, so as to prevent low-frequency noise from entering the fire detection device and improve the accuracy of fire detection.

(1) Obtaining spectral signal parameters of a light source of a site to be tested

In this embodiment, a light sensor is used to perform photoelectric conversion on a spectrum signal output by a spectrum filter, and send a converted analog electrical signal to an analog-to-digital converter, the analog electrical signal is converted into a digital signal by the analog-to-digital converter, and a first digital signal corresponding to a 5hz 4300 nm optical signal is used as a first spectrum signal parameter X1; taking a second digital signal corresponding to the optical signal with the wavelength of 5Hz and the wavelength of 3800 nm as a second spectral signal parameter X2; taking a third digital signal corresponding to the optical signal with the wavelength of 5Hz and the wavelength of 2700 nm as a third spectral signal parameter X3; taking a fourth digital signal corresponding to the 25Hz 4300 nm spectral signal as a fourth spectral signal parameter X4; taking a fifth digital signal corresponding to the 25Hz 3800 nm spectral signal as a fifth spectral signal parameter X5; the sixth digital signal corresponding to the 25hz 2700 nm spectral signal is taken as the sixth spectral signal parameter X6.

(2) Obtaining time signal parameters of a light source of a field to be measured

(3) Obtaining spatial signal parameters of a light source of a field to be measured

In the present embodiment, the first spectrum signal parameter X1 and the second spectrum signal are calculatedRatio of number parameter X2

And a third spatial signal parameter X12 is obtained.

Step S403: and substituting the spectral signal parameter, the time signal parameter and the space signal parameter of the light source of the field to be detected into a neural network model for fire detection to obtain the calculation output parameter of the light source of the field to be detected.

In this embodiment, the first to sixth spectral signal parameters X1 to X6, the first to third temporal signal parameters X7 to X9, and the first to third spatial signal parameters X10 to X12 may be input into the neural network model as input parameters, each set of input parameters corresponds to a calculation output parameter, and the output parameters may be used to represent the similarity of fire.

Step S404: and if the calculated output parameter of the light source of the field to be detected is greater than or equal to the preset output parameter threshold value, judging that the field to be detected is in fire.

Illustratively, as can be seen from the data analysis in table 1, the values of the output parameters of the neural network model may be 0,0.5,1,1.5,2,2.5 and 3, and thus, the preset output parameter threshold may be set to 2, that is, if the value of the calculated output parameter of the neural network model is greater than or equal to 2, it is determined that a fire occurs.

Optionally, the fire detection method further comprises the steps of: adding the spectral signal parameter, the time signal parameter, the spatial signal parameter and the output parameter of the light source of the field to be tested into a training data set to obtain an updated training data set; and correcting the neural network model according to the updated training data set.

In this embodiment, the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and the corresponding calculation output parameter of the light source in the field to be detected may be selectively stored in the memory of the fire detection device or the cloud server, the training test set is updated according to the data set of the light source in the field to be detected, and the neural network model is optimized according to the updated training test set, so that the false alarm rate may be reduced.

Therefore, according to the fire detection method provided by the embodiment of the invention, the neural network model is optimized through the characteristic parameters of the training light source, the characteristic parameters of the light source of the field to be detected are input into the neural network model for calculation, and fire identification is carried out according to the calculated output parameters, the identification algorithm is simple, the data processing amount is small, the response speed of fire detection is high, the accuracy is high, the requirement on hardware performance is reduced, the data analysis of the neural network model can be realized by using a smaller microprocessor, the practicability is strong, meanwhile, the neural network model can be learned and updated for the data of the field to be detected, and the false alarm rate is reduced.

Example two

The fire detection device provided by the embodiment of the invention can execute the fire detection method provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method. Fig. 6 is a schematic structural diagram of a fire detection device based on a neural network according to a second embodiment of the present invention.

As shown in fig. 6, a fire detection apparatus 100 according to an embodiment of the present invention includes: a signal processing unit 110 and a control unit 120.

The signal processing unit 110 includes a filtering unit 101, an optical sensor 102 and an analog-to-digital converter 103, where the filtering unit 101 is configured to receive an incident light signal and perform filtering processing on the incident light signal to obtain optical signals with at least two preset wavelengths; the optical sensor 102 is used for converting an optical signal into an analog electrical signal; the analog-to-digital converter 103 is used for converting the analog electrical signal into a digital signal; the control unit 120 is configured to analyze an incident light signal of the training light source, obtain a spectral signal parameter, a temporal signal parameter, and a spatial signal parameter of the training light source, establish a training data set according to the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and a preset output parameter of the training light source, and establish a neural network model for fire detection according to the training data set; the control unit 120 is further configured to determine a calculation output parameter of the site to be tested according to the neural network model, and determine whether a fire occurs in the site to be tested according to the calculation output parameter.

The training light source comprises flame, an interference light source and flame arranged under the interference light source, and the interference light source comprises an arc light source, a heater light source and a lamplight source.

Optionally, the light sensor 102 includes an infrared light sensor and an ultraviolet light sensor.

For example, an infrared light filtering unit with three wavelengths may be disposed on the surface of the light sensor to filter the incident light signal, or an infrared light filtering unit and an ultraviolet light filtering unit with three wavelengths may be disposed on the surface of the light sensor to filter the incident light signal, wherein the ultraviolet light sensor may be used to filter ultraviolet light with a wavelength of 160 nm to 220 nm, and may be used to identify the interfering light source arc welding ultraviolet light, thereby improving the accuracy of fire identification.

Optionally, as shown in fig. 6, the fire detection apparatus 100 further includes: the cloud server 130, the cloud server 130 is used for storing the training data set.

In this embodiment, the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and the output parameter of the light source in the field to be detected may be selectively stored in the cloud server of the fire detection apparatus, the training test set is updated according to the data set of the light source in the field to be detected, and the neural network model is optimized according to the updated training test set, so that the false alarm rate may be reduced.

Further, the control unit 120 includes: the device comprises a spectral signal parameter acquisition unit, a time signal parameter acquisition unit and a spatial signal parameter acquisition unit, wherein the spectral signal parameter acquisition unit is used for acquiring spectral signal parameters of a training light source according to spectral intensities corresponding to optical signals with various wavelengths; the time signal parameter acquisition unit is used for acquiring a plurality of spectral signal parameters, and respectively carrying out derivation operation on each spectral signal parameter to acquire a time signal parameter of the training light source; the spatial signal parameter acquiring unit is used for acquiring spectral signal parameters corresponding to a plurality of optical signals with different preset wavelengths, calculating a ratio between the spectral signal parameters and acquiring spatial signal parameters of the training light source.

Further, the time signal parameter obtaining unit is further configured to obtain a first spectral signal parameter corresponding to the optical signal with the first preset frequency and the first preset wavelength, a second spectral signal parameter corresponding to the optical signal with the first preset frequency and the second preset wavelength, and a third spectral signal parameter corresponding to the optical signal with the first preset frequency and the third preset wavelength, respectively, where the first preset wavelength, the second preset wavelength, and the third preset wavelength may be adjusted according to the fuel type of the site to be measured; and respectively calculating time derivatives of the first spectral signal parameter, the second spectral signal parameter and the third spectral signal parameter to obtain a first time parameter, a second time parameter and a third time parameter.

Further, the spatial signal parameter obtaining unit is further configured to obtain a first spectral signal parameter corresponding to the optical signal with the first preset frequency and the first preset wavelength, a second spectral signal parameter corresponding to the optical signal with the first preset frequency and the second preset wavelength, and a third spectral signal parameter corresponding to the optical signal with the first preset frequency and the third preset wavelength, respectively, where the first preset wavelength, the second preset wavelength, and the third preset wavelength may be adjusted according to the fuel type of the site to be measured; calculating the ratio of the first spectral signal parameter to the second spectral signal parameter to obtain a first spatial signal parameter; calculating the ratio of the first spectral signal parameter to the third spectral signal parameter to obtain a second spatial signal parameter; and calculating the average value of the ratio of the first spectral signal parameter, the third spectral signal parameter and the second spectral signal parameter to obtain a third spatial signal parameter.

Further, the control unit 120 further includes: the neural network model acquisition unit is used for establishing a neural network model, and the neural network model comprises a weight matrix and a deviation matrix; the training data set acquisition unit is used for taking the spectral signal parameter, the time signal parameter and the spatial signal parameter of each training light source as input parameters of the neural network model, acquiring preset output parameters of each training light source, and establishing a training data set according to all the input parameters and the corresponding preset output parameters; the neural network model optimization unit is used for carrying out iterative training on the neural network model according to the training data set, constructing a loss function, detecting the value of the loss function once every iteration, acquiring a converged weight matrix and a converged deviation matrix if the value of the loss function is smaller than a preset convergence threshold value, and constructing the neural network model for fire detection according to the converged weight matrix and the converged deviation matrix; and the fire detection unit is used for substituting the spectral signal parameter, the time signal parameter and the space signal parameter of the light source of the field to be detected into the neural network model for fire detection to obtain the calculation output parameter of the light source of the field to be detected, and if the calculation output parameter is greater than or equal to a preset output parameter threshold value, judging that the field to be detected has a fire.

Further, the control unit 120 is further configured to use a flame generated by burning the fuel in the site to be tested in an environment without an interference light source as a supplementary training light source, where a fuel type of the supplementary training light source is different from a fuel type of the training light source; the control unit 120 is further configured to obtain incident light signals of a plurality of supplementary training light sources, perform signal processing and analysis on the incident light signals of each supplementary training light source, and obtain spectral signal parameters, temporal signal parameters, and spatial signal parameters of the supplementary training light sources; the control unit 120 is further configured to establish a supplementary training data set according to the spectral signal parameter, the temporal signal parameter, the spatial signal parameter, and the corresponding supplementary preset output parameter of the supplementary training light source; the control unit 120 is further configured to optimize the neural network model based on the supplemental training data set and the training data set.

Further, the control unit 120 further includes: the storage unit is used for adding the spectral signal parameters, the time signal parameters, the spatial signal parameters and the corresponding calculation output parameters of the light source of the field to be tested to the training data set to obtain an updated training data set; and the neural network model correction unit is used for correcting the neural network model according to the updated training data set.

Therefore, the fire detection device provided by the embodiment of the invention optimizes the neural network model by training the characteristic parameters of the light source, inputs the characteristic parameters of the light source of the field to be detected into the neural network model for calculation, and carries out fire identification according to the calculated output parameters, the identification algorithm is simple, the data processing amount is small, the response speed of fire detection is high, the accuracy is high, the requirement on hardware performance is reduced, the data analysis of the neural network model can be realized by using a smaller microprocessor, the practicability is strong, meanwhile, the neural network model can be learned and updated for the data of the field to be detected, and the false alarm rate is reduced.

EXAMPLE III

Fig. 7 is a schematic structural diagram of an apparatus according to a third embodiment of the present invention. Fig. 7 illustrates a block diagram of an exemplary device 12 suitable for use in implementing embodiments of the present invention. The device 12 shown in fig. 7 is only an example and should not bring any limitation to the function and scope of use of the embodiments of the present invention.

As shown in FIG. 7, device 12 is in the form of a general purpose computing device. The components of device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28 for storing one or more programs; the bus 18, which connects the various system components (including the system memory 28 and the processing unit 16), when one or more programs are executed by the one or more processors, causes the one or more processors to implement the fire detection method of embodiments of the present invention.

Bus 18 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, such architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, micro-channel architecture (MAC) bus, enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

Device 12 typically includes a variety of computer system readable media. Such media may be any available media that is accessible by device 12 and includes both volatile and nonvolatile media, removable and non-removable media.

The system memory 28 may include computer system readable media in the form of volatile memory, such as Random Access Memory (RAM) 30 and/or cache memory 32. Device 12 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 34 may be used to read from and write to non-removable, nonvolatile magnetic media (not shown in FIG. 7, and commonly referred to as a "hard drive"). Although not shown in FIG. 7, a magnetic disk drive for reading from and writing to a removable, nonvolatile magnetic disk (e.g., a "floppy disk") and an optical disk drive for reading from or writing to a removable, nonvolatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 by one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to carry out the functions of embodiments of the invention.

A program/utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which examples or some combination thereof may comprise an implementation of a network environment. Program modules 42 generally carry out the functions and/or methodologies of the described embodiments of the invention.

Device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), with one or more devices that enable a user to interact with device 12, and/or with any devices (e.g., network card, modem, etc.) that enable device 12 to communicate with one or more other computing devices. Such communication may be through an input/output (I/O) interface 22. Also, the device 12 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via the network adapter 20. As shown, the network adapter 20 communicates with the other modules of the device 12 via the bus 18. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

The processing unit 16 executes various functional applications and data processing, such as implementing the fire detection method provided by embodiments of the present invention, by executing programs stored in the system memory 28.

Example four

The fourth embodiment of the present invention further provides a storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the fire detection method.

Storage media for embodiments of the present invention may take the form of any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + +, or the like, as well as conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).

It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments illustrated herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A fire detection method based on a neural network is characterized by comprising the following steps:

acquiring a plurality of training light sources, wherein each training light source comprises a flame, an interference light source and a flame arranged under the interference light source;

adjusting the distance between flames in the training light sources and a flame detection device, acquiring incident light signals of a plurality of training light sources, performing signal processing and analysis on the incident light signals of each training light source, and acquiring spectral signal parameters, time signal parameters and spatial signal parameters of each training light source, wherein the spectral signal parameters represent spectral intensities corresponding to optical signals with multiple wavelengths, the time signal parameters represent the trend of the spectral signal parameters corresponding to the optical signals with the same wavelength changing along with time, and the spatial signal parameters represent the trend of relative change of the spectral signal parameters corresponding to the optical signals with different wavelengths;

establishing a training data set according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and the corresponding preset output parameter of the training light source;

constructing a neural network model for fire detection according to the training data set;

and determining a calculation output parameter of a light source of the field to be detected according to the neural network model, and determining whether the field to be detected has a fire or not according to the calculation output parameter.

2. The fire detection method based on the neural network as claimed in claim 1, wherein the steps of obtaining the incident light signal of each of the training light sources, performing signal processing and analysis on each of the light signals, and obtaining the spectral signal parameter, the temporal signal parameter and the spatial signal parameter of each of the training light sources comprise:

acquiring at least one optical signal with preset frequency sent by each training light source;

filtering each incident light signal with the preset frequency to obtain light signals with at least two preset wavelengths;

performing photoelectric conversion and analog-to-digital conversion processing on each optical signal with the preset wavelength to obtain spectral signal parameters of the training light source;

acquiring a plurality of spectral signal parameters, and respectively performing derivation operation on each spectral signal parameter to acquire a time signal parameter of the training light source;

acquiring the spectral signal parameters corresponding to a plurality of optical signals with different preset wavelengths, calculating the ratio of each spectral signal parameter, and acquiring the spatial signal parameters of the training light source.

3. The fire detection method based on the neural network as claimed in claim 2, wherein the step of obtaining a plurality of spectral signal parameters, respectively performing a derivation operation on each spectral signal parameter to obtain a time signal parameter of the training light source comprises the steps of:

respectively obtaining a first spectral signal parameter corresponding to an optical signal with a first preset frequency and a first preset wavelength, a second spectral signal parameter corresponding to an optical signal with a first preset frequency and a second preset wavelength, and a third spectral signal parameter corresponding to an optical signal with a first preset frequency and a third preset wavelength, wherein the first preset wavelength, the second preset wavelength and the third preset wavelength can be adjusted according to the fuel type of a field to be measured;

and respectively calculating time derivatives of the first spectral signal parameter, the second spectral signal parameter and the third spectral signal parameter to obtain a first time parameter, a second time parameter and a third time parameter.

4. The fire detection method based on the neural network as claimed in claim 2, wherein the steps of obtaining the spectral signal parameters corresponding to a plurality of optical signals with different preset wavelengths, calculating the ratio between each of the spectral signal parameters, and obtaining the spatial signal parameter of the training light source comprise:

calculating the ratio of the first spectral signal parameter to the second spectral signal parameter to obtain a first spatial signal parameter;

calculating the ratio of the first spectral signal parameter to the third spectral signal parameter to obtain a second spatial signal parameter;

and calculating the average value of the ratio of the first spectral signal parameter, the third spectral signal parameter and the second spectral signal parameter to obtain a third spatial signal parameter.

5. The neural network-based fire detection method of claim 1, further comprising, after constructing a neural network model for fire detection from the training data set, the steps of:

taking flame generated by burning fuel of a field to be tested in an interference-free light source environment as a supplementary training light source, wherein the fuel type of the supplementary training light source is different from that of the training light source;

adjusting the distance between the supplementary training light source and the flame detection device, acquiring a plurality of incident light signals of the supplementary training light source, and performing signal processing and analysis on the incident light signals of each supplementary training light source to acquire spectral signal parameters, time signal parameters and space signal parameters of the supplementary training light source;

establishing a supplementary training data set according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and the corresponding supplementary preset output parameter of the supplementary training light source;

and optimizing a neural network model according to the supplementary training data set and the training data set.

6. The neural network-based fire detection method of claim 1, wherein the building of the neural network model for fire detection from the training data set comprises the steps of:

establishing a neural network model, wherein the neural network model comprises a weight matrix and a deviation matrix;

performing iterative training on the neural network model according to the training data set, constructing a loss function, detecting the value of the loss function once every iteration, and if the value of the loss function is smaller than a preset convergence threshold value, acquiring a converged weight matrix and a converged deviation matrix;

and constructing a neural network model for fire detection according to the converged weight matrix and the converged deviation matrix.

7. The neural network-based fire detection method of claim 1, further comprising the steps of:

acquiring an incident light signal of a light source of a field to be detected;

analyzing an incident light signal of the light source of the field to be detected to obtain a spectrum signal parameter, a time signal parameter and a space signal parameter of the light source of the field to be detected;

substituting the spectral signal parameter, the time signal parameter and the space signal parameter of the light source of the field to be detected into the neural network model for fire detection to obtain the calculation output parameter of the light source of the field to be detected;

adding the spectral signal parameter, the time signal parameter and the spatial signal parameter of the light source of the field to be tested and the corresponding calculation output parameter to the training data set to obtain an updated training data set;

and correcting the neural network model according to the updated training data set.

8. A fire detection device based on a neural network, comprising: a signal processing unit and a control unit, wherein,

the signal processing unit comprises a filtering unit, an optical sensor and an analog-to-digital converter, wherein the filtering unit is used for receiving incident optical signals and filtering the incident optical signals to obtain optical signals with at least two preset wavelengths, the optical sensor is used for converting the optical signals into analog electric signals, and the analog-to-digital converter is used for converting the analog electric signals into digital signals;

the control unit is used for analyzing an incident light signal of the training light source, and acquiring a spectral signal parameter, a time signal parameter and a spatial signal parameter of the training light source, wherein the spectral signal parameter represents spectral intensities corresponding to optical signals with multiple wavelengths, the time signal parameter represents a trend of the spectral signal parameter corresponding to the optical signal with the same wavelength changing along with time, the spatial signal parameter represents a trend of relative change of the spectral signal parameter corresponding to the optical signal with different wavelengths, a training data set is established according to the spectral signal parameter, the time signal parameter, the spatial signal parameter and corresponding preset output parameters of the training light source, and a neural network model for fire detection is established according to the training data set;

the control unit is also used for determining the calculation output parameters of the field to be detected according to the neural network model and determining whether the field to be detected has a fire or not according to the calculation output parameters.

9. An apparatus, comprising:

one or more processors;

a system memory to store one or more programs;

when executed by the one or more processors, cause the one or more processors to implement the neural network-based fire detection method of any one of claims 1-7.

10. A storage medium having stored thereon a computer program, characterized in that the program, when being executed by a processor, implements the neural network-based fire detection method as claimed in any one of claims 1 to 7.