CN102743830A - Automatic electric switch cabinet fire extinguishing system and fire recognition method - Google Patents
Automatic electric switch cabinet fire extinguishing system and fire recognition method Download PDFInfo
- Publication number
- CN102743830A CN102743830A CN201210237440XA CN201210237440A CN102743830A CN 102743830 A CN102743830 A CN 102743830A CN 201210237440X A CN201210237440X A CN 201210237440XA CN 201210237440 A CN201210237440 A CN 201210237440A CN 102743830 A CN102743830 A CN 102743830A
- Authority
- CN
- China
- Prior art keywords
- flame
- pixel
- fire
- value
- fire extinguishing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention discloses an automatic electric switch cabinet fire extinguishing system and a fire recognition method, wherein the automatic electric switch cabinet fire extinguishing system comprises a fire extinguishing agent nozzle which is arranged in an electric switch cabinet; the fire extinguishing agent nozzle is connected with a storage device in which a non-conducting fire extinguishing agent is stored through a pipeline; an electromagnetic valve which is controlled by a central controller is arranged on a pipeline; a photoelectric sensor is also arranged in the electric switch cabinet; and the photoelectric sensor outputs a detection signal to the central controller and then the central controller judges whether to open the electromagnetic valve. The automatic electric switch cabinet fire extinguishing system is capable of responding sudden electric fire and reducing the loss caused thereby. Fire extinguishing agents such as heptafluoropropane and superfine dry powder are used and non-conducting, so that secondary danger caused by conduction of fire extinguishing agents does not occur; and the fire extinguishing agents are environmental-friendly, so that the damage to the environment caused by conduction of fire extinguishing agents also does not occur. The photoelectric sensor is used, so that the fire such as electric arc and the like in the switch cabinet can be rapidly and correctly detected.
Description
Technical field
The invention belongs to the electrical switchgear technical field of security and protection, relate to a kind of electrical switchgear automatic fire extinguishing system and fire recognition methods.
Background technology
Switch cubicle is a kind of electric equipment, and outside line is introduced into master switch in the cabinet, gets into then to divide the control switch, and each needs to be provided with by it along separate routes.Like instrument, automatic control, the motor magnetic switch, various A.C. contactors etc., what have also establishes hyperbaric chamber and low-pressure chamber switch cubicle, is provided with high voltage bus, like power plant etc.
Switch cubicle divides high-tension switch cabinet and low-tension switch cabinet:
High-tension switch cabinet is meant and is used for power system generating, transmission of electricity, distribution, electric energy conversion and has consumed effects such as break-make, control or protection; Electric pressure is in the electric equipment products of 3.6kV ~ 550kV; Comprise that mainly primary cut-out, high voltage isolator overlap and sectionaliser with earthed switch, high voltage load switch, high pressure automatically, several big type of high-voltage actuating mechanism, high-voltage explosion-proof power distribution equipment and high-tension switch cabinet etc.High-tension switch cabinet should satisfy the relevant requirements of GB3906-1991 " 3-35kV ac metal closing switch gear " standard; Are made up of cabinet and breaker two major parts, cabinet is made up of housing, electric elements (comprising insulating part), various mechanism, secondary terminal and line etc.
Low-tension switch cabinet is applicable to industries such as power plant, oil, chemical industry, metallurgy, weaving, skyscraper, as the usefulness of transmission of electricity, distribution and electric energy conversion.Product meets IEC439-1, the standard code of GB7251.1-1997 " low-voltage complete switch equipment ".
Switch cubicle is an elementary cell in the network system, the executing agency of electric power system control.Along with unattended operation transformer station, intelligent grid The Application of Technology, it is particularly important that the security protection of switch cubicle becomes.
Switch cubicle owing to reasons such as long-play, environmental change, line load disturbances, disasters such as localized hyperthermia, electric arc, burning can occur in the course of the work, and these disasters can cause cabinet badly damaged, and electric power system is destroyed and personnel's injury.Therefore be necessary to study the security protection of switch cubicle, guarantee executing agency's operate as normal, loss is reduced to bottom line, to improve power supplying efficiency.
The security protection of switch cubicle mainly is to prevent high temperature, burning (naked light, glow); Guarantee the switch operate as normal, or when equipment component breaks down, avoid other equipment to be damaged, loss is reduced to minimum level; As find the switch cubicle burning, put out a fire automatically; Find that the switch cubicle temperature is too high, report to the police etc.
Along with the progress of society, people's awareness of the importance of fire prevention is more and more stronger, and the demand of automatic fire extinguishing system more and more widely.Automatic fire extinguishing system is the fixed fire extinguishing system of commonplace use.Safe and reliable, economical and practical, fire-fighting efficiency advantages of higher that this system has.
The automatic fire extinguishing system type that adopts at present both at home and abroad is more, like automatic wet pipe sprinkler system, Pre-Action Automatic Sprinkling Fire System, dry type and preact associating automatic sprinkler system, dry automatic sprinkler system, the automatic sprinkler system that drenches with rain, water spray fire extinguish system etc.Automatic fire extinguishing system not only generally uses in skyscraper, public building, factory and warehouse, and has developed into automatic sprinkler system is installed in the home dwelling.
But electric complete set of equipments switch cubicle is different from above-mentioned category, behind electric elements or electric wire, the ignition of cable, can not directly fire-fight by water in the switch cubicle.Because contain the impurity of conduction in the water, be sprayed on the charging equipment dust impurity on the infiltration equipment again, then easy conductive more.As fire-fight by water, also can reduce the insulating properties of electric equipment, cause ground short circuit, or jeopardize near fire fighting personnel safety.
Summary of the invention
The technical problem that the present invention solves is to provide a kind of cabinet automatic fire extinguishing system; Overcoming direct water, having on the basis of the problem of polluting the extinguishing chemical fire extinguishing; Through naked light and the electric arc in the photodetector search switch cabinet, improve the recognition capability and the automatic processing capabilities of right fire.
The present invention realizes through following technical scheme:
A kind of cabinet automatic fire extinguishing system comprises the extinguishing chemical shower nozzle that is installed in the cabinet, and the extinguishing chemical shower nozzle has the storage of non-conducting extinguishing agent to be connected through pipeline with storage, and pipeline is provided with the magnetic valve by central controller controls;
Also be provided with photoelectric sensor in the cabinet; Photoelectric sensor comprises pick-up lens and optical-electrical converter; Pick-up lens is provided with infrared fileter, and pick-up lens is taken the formation black white image and converted image digital signal into through optical-electrical converter, and the white portion in the black white image is a flame; Black part is divided into background, and photoelectric sensor outputs to central controller with detection signal;
Central controller is discerned fire according to flame behavioral characteristics and static nature in the middle of the detection signal, and behavioral characteristics characterizes the frequency of flashing of flame, and static nature characterizes the irregularly shaped of flame, differentiates C=P with circularity
2/ 4 π S, wherein C is a circularity, and P is a girth, and S is an area; Flash frequency and during greater than 1.8 illuminator, determine that it is fire at 7～12HZ, circularity when detecting, the central controller controls magnetic valve is opened the release non-conducting extinguishing agent.
Described extinguishing chemical shower nozzle is provided with a plurality of, is connected with storage through the pipeline that is provided with magnetic valve respectively.
Described storage is a pressure storage type extinguishing chemical storage, and its pressure is 1～4Mpa;
Described non-conducting extinguishing agent is heptafluoro-propane or ultra-fine dry powder extinguishing agent.
Also be provided with the hygrosensor that is connected with central controller controls in the described cabinet.
Described hygrosensor is fibre optic temperature sensor or infrared temperature sensor, and fibre optic temperature sensor is arranged in the cabinet with contactless mode with mode, the infrared temperature sensor of contact.
A kind of fire recognition methods of cabinet automatic fire extinguishing system may further comprise the steps:
1) is arranged on the photoelectric sensor photographic images in the cabinet and transmit signal; The pick-up lens of photoelectric sensor is provided with infrared fileter; Pick-up lens is taken the formation black white image and is converted image digital signal into through optical-electrical converter; White portion in the black white image is a flame, and black part is divided into background;
2) central controller receives detection signal; Image digital signal is carried out extracting static nature and the behavioral characteristics that flame is different from optical jammer source after the preliminary treatment; And it is vectorial that it is quantized to form Fire Criterion, gets rid of optical jammer source and discern fire through analyzing the criterion vector;
Described static nature extracts through the single-frame images computing and extracts the target signature that comprises wedge angle, circularity and trapezoidal characteristics:
The wedge angle characteristic: some wedge angles up can appear in flame fringe, and the continuous variation of flame wedge angle number is the performance of flame fringe shake, adopt based target edge extracting wedge angle number;
The circularity characteristic: the shape of flame is irregular, describes the regularity of its shape through circularity, circularity be calculated as C=P
2/ 4 π S, wherein C is a circularity, and P is a girth, and S is an area;
Trapezoidal characteristics: flame base is near comburant, and the upper width that makes flame is described this characteristic less than bottom width through the distance of regional barycenter and central longitudinal coordinate;
Behavioral characteristics be through the multiple image computing extract comprise beat, target signature that the edge enlarges characteristic:
The characteristic of beating: visually, flame flicking shows as flame and ceaselessly beats, and extracts the height of target in 25 two field pictures; Be designated as x [n] (n=0,1,2 ... 24); Draw frame number-height oscillogram according to x [n], then the crest number is counted, have only when crest value during than all big 4 pixels tall of the trough value of both sides; Just carry out frequency counting, the crest number is designated as the target jumping frequency rate;
The edge enlarges characteristic: at the fire early period of origination, the flame main body increases gradually, and the girth through the target area increases gradually describes this characteristic;
3) obtain the characteristic vector data of flame after, these characteristic synthetics are handled: earlier the sample characteristics vector is input to the BP neutral net; Output layer output valve and ideal value are relatively obtained output error; According to the output error backpropagation, successively calculate the error of each unit of latent layer forward, and with these error correction anterior layer weights; So repeat repeatedly the process of input of sample information forward and error back propagation, be performed until output error and reach the acceptable degree, or reach till the predefined study number of times; Input layer is done normalization and is handled for the characteristic vector of the infrared light supply of extraction, the output valve of output layer, field condition is made the judgement of fire condition, precarious position and normal condition according to output valve.
Described preliminary treatment to image digital signal comprises:
1) binary conversion treatment, handle according to following binaryzation algorithm:
Wherein, Binary is a binary map pixel gray value; F is a gray-scale map pixel gray value; T is the binary conversion treatment threshold value, makes T=155;
2) medium filtering: target pixel points and 2n neighborhood territory pixel point gray value are carried out descending ordering; N is a positive integer; Get n+1 value and compose and give object pixel, if having in object pixel and the 2n neighborhood territory pixel gray value less than n+1 individual be 255, the object pixel gray value is changed to 0; Having more than or equal to n+1 is 255, and the object pixel gray value is changed to 255;
3) target label:, be different gray values with different bright connected component labelings for distinguishing different bright connected regions:
A, the traversal bianry image, with first object pixel that traverses as sub pixel;
B, with the serve as a mark input parameter of function of sub pixel coordinate, with this point of gray value mark of an appointment;
C, travel through 8 neighborhoods of seed points one by one, if object pixel stops traversal, invocation flags function self;
After d, intact this connected region of mark, turn back to a step, adopting uses the same method continues other connected region of mark;
4) hole is filled: when the inner local luminance of flame was low, some holes can appear in the flame body interior after binary conversion treatment, will fill these holes, obtaining a complete flame main body, were filled to based on the hole of background area mark:
A, traversing graph picture are that 0 pixel is as sub pixel with a gray value that finds;
B, all pixels of sub pixel place connected region are labeled as Lab [n] (n=0,1,2 through non-recursive algorithm ...);
C, continuing the traversing graph picture, is 0 pixel if traverse a gray value, and n=n+1 gets back to the b step;
After d, view picture figure traveled through completion, background pixel was marked as Lab [0], and hole is by distinguishing mark, was that the grey scale pixel value of L [0] is changed to 0 with mark value, and the grey scale pixel value of non-L [0] is changed to 255;
5) revise mark: with the 0th frame is basic frame, judge in other frame each boundary marker value whether with the 0th frame same position on the boundary marker value consistent, if inconsistent, with its be revised as with the 0th frame in boundary marker value on the same position.
Described wedge angle Feature Extraction is based on object edge and extracts the wedge angle number, utilizes recurrence method and backtracking method target-marking border, and writes down the coordinate sign indicating number of each boundary point; Judge according to the coordinate chain code whether each object boundary point is the summit then, and can constitute wedge angle with left and right sides neighborhood;
The wedge angle of flame fringe is similar to triangle, and the summit is exactly the maximum point of flame fringe y coordinate, according to the coordinate chain code that travels through object boundary one by one, relatively its y coordinate whether than about the y coordinate figure of 5,10,15 pixels all big; If satisfy condition, think that then these pixels constitute a wedge angle;
When object edge generation small variations; To judge also whether the height of wedge angle meets the characteristic of flame wedge angle; After the wedge angle summit is confirmed; Line between summit, the 15th boundary pixel point of the 15th the boundary pixel point in the left side, summit and the right constitutes a triangle, this leg-of-mutton height is designated as the height of wedge angle;
Triangle length of side a, b, c can obtain through range formula:
Ask leg-of-mutton area according to Heron's formula again:
q=(a+b+c)/2 （4）
Area formula S=c*h/2 substitution formula (3) is got:
When h surpasses preset threshold, the wedge angle number of this target is added 1, and wedge angle summit grey scale pixel value is changed to 255; Need jump to 16th boundary pixel point of this summit back and proceed judge when traveling through boundary pixel this moment;
During described circularity feature extraction, girth P is calculated as:
1. adopt recurrence method and backtracking method target-marking border;
2. Freeman 8 direction chain codes coding is carried out on the border;
3. girth is initialized as 0; If chain code is an odd number; Girth adds 1, otherwise girth adds
Area S is calculated as: utilize each bright area of onrecurrent region growing method mark, traversal view picture figure asks the area of the corresponding number of pixels of different mark value as each marked region;
Being calculated as of the distance of the regional barycenter of trapezoidal characteristics and central longitudinal coordinate:
The ordinate of regional barycenter does
Wherein gy is the center of gravity ordinate, and S is the target area area, and R is the target area, and y is the ordinate of each pixel in the target area;
Classification and boundary coordinate according to boundary point calculate the center of gravity ordinate:
A, utilize target area computational methods reference area S based on the boundary point classification;
B, make
initialization Y=0;
If c, traversal object boundary point classification code are boundary point (x
n, y
n) be the lower boundary point,
If boundary point (x
n, y
n) be the coboundary point,
If boundary point (x
n, y
n) be the horizontal direction summit, Y=Y+y
n(n=0,1,2 wherein ... H-1, the height of h representative image.）
D, the final Y result of Y is updated to (7) formula, obtains the center of gravity ordinate;
E, according to the minimum and maximum value of object boundary point ordinate, calculate the ordinate y of target's center
c:
y
min=Min(y)；y
max=Max(y)；y∈R;
Adopt the backpropagation of Sigmoid type output function to be:
(1) selected weight coefficient initial value;
(2) repeat following process until convergence, with sample characteristics amount fan-in network successively:
1. calculate each layer unit output o from forward direction
j
2. calculate each elemental error δ of output layer
j
δ
j=(y-o
j)o
j(1-o
j) （11）
3. the reverse transfer error is calculated each latent each elemental error δ of layer
j
4. calculate the correction of each layer weights
△ω
ij(t)=α△ω
ij(t-1)+ηδ
jo
j （13）
5. according to each layer of each layer weights correction correction weights
ω
ij(t+1)=ω
ij(t)+△ω
ij(t) （14）
When adopting three layers of feedforward network to carry out the branch time-like of fire condition, precarious position and normal condition, the input layer number is the dimension of characteristic vector; Hide the node layer number and round up for
, wherein p is the input layer number; Output layer node number is a number of categories; In the forward process, the weights initial value chooses usually ± 0.3 interval random number; In the reverse transfer method, step-length η directly sounds out 0.1～3; Inertia item alpha is selected between 0.9～1.
Compared with prior art, the present invention has following beneficial technical effects:
Cabinet automatic fire extinguishing system provided by the invention; Through the situation timing monitoring of photodetector to switch cubicle; Judge through central controller, and non-conducting extinguishing agent is opened back ejection fire extinguishing through shower nozzle at magnetic valve, has realized the self-extinguishing of cabinet fully.Be particularly suitable for harsh and unforgiving environments, and volume is little, does not influence the normal operation of equipment under test like power plant, power supply department transformer station, unattended operation transformer station, and easy installation and reliable.
Cabinet automatic fire extinguishing system provided by the invention has adopted heptafluoro-propane or ultra-fine dry powder extinguishing agent, and these extinguishing chemicals are nonconducting, therefore can be owing to the extinguishing chemical conduction causes secondary harm; And a little extinguishing chemicals are environmental protection, therefore can not lead environment is worked the mischief owing to extinguishing chemical.
Heptafluoro-propane is gaseous state at normal temperatures, and colorless and odorless, non-conductively have good gas phase electrical insulating property, do not have corrosion, no environmental protection restriction, it is shorter that atmosphere retains the phase, has good spatter property (in atmosphere, vaporize fully and do not stay residue).Extinguishing mechanism mainly is to interrupt pyric chain, and blow-off velocity is exceedingly fast, and this is favourable to rescue property protection sophisticated electronics and valuables.The non-toxic reaction of heptafluoro-propane (NOAEL) concentration is 9%, and toxic reaction (LOAEL) concentration is 10.5%, and the design concentration of heptafluoro-propane is generally less than 10%, to human body safety.Its characteristics and the good physical property that is applicable to the fire extinguishing system use.
Ultra-fine dry powder extinguishing agent, nontoxic, pollution-free, do not destroy atmospheric ozone layer, the protection thing is not had corrosion, non-stimulated to human body skin and respiratory tract, fire extinguishing rapidly, the efficient height.
Cabinet automatic fire extinguishing system provided by the invention adopts photoelectric sensor, can fast, accurately detect the fire in the switch cubicle, like electric arc etc.Reflect the monitoring field condition through detector intuitively through image energy, just can discern fire automatically, the record fire at the fire early period of origination.And monitoring range is big, and this is that the conventional fire detector can't be accomplished.
Cabinet automatic fire extinguishing system provided by the invention is fit to various switch cubicles and installs and uses: both can be used for novel switched cabinet and also can be used for the switch cubicle that using simultaneously; Also further can a plurality of switch cubicle extinguishing devices be networked, possess the remote monitoring function.
Description of drawings
Fig. 1 is one of structural representation of the present invention;
Fig. 2 is two of a structural representation of the present invention;
Fig. 3 is two of a structural representation of the present invention;
Wherein: 1 is cabinet; 2 is central controller; 3 is photoelectric sensor; 4 is storage; 5 is magnetic valve; 6 is pipeline; 7 is the extinguishing chemical shower nozzle;
Fig. 4 is the image process chart of detection signal;
Fig. 5 is image pretreatment process figure;
Fig. 6 is object pixel 8 neighborhood traversal order;
Fig. 7 is image characteristics extraction figure;
Fig. 8 is a wedge angle feature extraction sketch map;
Fig. 9 is Freeman 8 direction chain code sketch mapes.
The specific embodiment
Below in conjunction with accompanying drawing the present invention is done to describe in further detail, said is to explanation of the present invention rather than qualification.
As shown in Figure 1; A kind of cabinet automatic fire extinguishing system; Comprise the extinguishing chemical shower nozzle 7 that is installed in the cabinet 1, extinguishing chemical shower nozzle 7 has the storage 4 of non-conducting extinguishing agent to be connected through pipeline 6 with storage, and pipeline 6 is provided with the magnetic valve 5 by central controller 2 controls; Also be provided with photoelectric sensor 3 in the cabinet 1, photoelectric sensor 3 outputs to central controller 2 with detection signal, judges whether open magnetic valve 5 by it.
Concrete photoelectric sensor 3 comprises pick-up lens and optical-electrical converter; Pick-up lens is provided with infrared fileter; Pick-up lens is taken the formation black white image and is converted image digital signal into through optical-electrical converter; White portion in the black white image is a flame, and black part is divided into background, and photoelectric sensor outputs to central controller with detection signal;
Central controller is discerned fire according to flame behavioral characteristics and static nature in the middle of the detection signal, and behavioral characteristics characterizes the frequency of flashing of flame, and static nature characterizes the irregularly shaped of flame, differentiates C=P with circularity
2/ 4 π S, wherein C is a circularity, and P is a girth, and S is an area; Flash frequency and during greater than 1.8 illuminator, determine that it is fire at 7～12HZ, circularity when detecting, the central controller controls magnetic valve is opened the release non-conducting extinguishing agent.
Like Fig. 2, shown in Figure 3, for the different levels (twice or three layers) or the different requirement in space separately that adapt to switch cubicle, the space for each opposition in the cabinet 1 is equipped with independently photoelectric sensor 2; And dispose an extinguishing chemical shower nozzle 7; Each photoelectric sensor 3 all is connected with central controller 2, it is implemented monitoring, in the time of needs; Central controller 2 sends instruction unpack extinguishing chemical shower nozzle 7, and extinguishing chemical shower nozzle 7 ejection non-conducting extinguishing agents are put out a fire.Accordingly, central controller need judge that the magnetic valve of controlling then on the corresponding pipeline is opened to the fire place when detecting the fire generation, and release fire suppression agent is put out a fire.
Concrete, pipeline 6 adopts withstand voltage high tube connector, in the position of needs fire extinguishing shower nozzle is installed, and shower nozzle is connected to one or more fire extinguishings position, guarantees that extinguishing chemical directly sprays ignition point, improves fire-fighting efficiency.
Because the specific (special) requirements of described cabinet, the concrete selection of said non-conducting extinguishing agent is extinguishing chemicals such as heptafluoro-propane, ultra-fine dry powder.
Heptafluoro-propane is gaseous state at normal temperatures, colorless and odorless, non-conductive, nothing corrosion, and no environmental protection restriction, it is shorter that atmosphere retains the phase.Extinguishing mechanism mainly is to interrupt pyric chain, and blow-off velocity is exceedingly fast, and this is favourable to rescue property protection sophisticated electronics and valuables.The non-toxic reaction of heptafluoro-propane (NOAEL) concentration is 9%, and toxic reaction (LOAEL) concentration is 10.5%, and the design concentration of heptafluoro-propane is generally less than 10%, to human body safety.Its characteristics have good spatter property (residue is not stayed in vaporization fully in atmosphere), good gas phase electrical insulating property and the good physical property that is applicable to the fire extinguishing system use.
Ultra-fine dry powder extinguishing agent is a kind of efficient and environment-friendly type fire product of new generation.ABC ultra-fine dry powder extinguishing agent commonly used is nontoxic, pollution-free, do not destroy atmospheric ozone layer, and the protection thing is not had corrosion, and is non-stimulated to human body skin and respiratory tract, fire extinguishing rapidly, the efficient height is the substitute of halon fire agent.
Be pressure storage type extinguishing chemical storage in the described storage, its pressure is for (1～4MPa) after opening under the control of magnetic valve at controller, makes that the extinguishing chemical in the storage directly is sprayed onto the fire extinguishing position through piping under the effect of himself pressure, realizes putting out a fire.
Further, also be provided with the hygrosensor that is connected with central controller controls in the described cabinet, hygrosensor is used for the temperature of critical component in the search switch cabinet; Hygrosensor is fibre optic temperature sensor or infrared temperature sensor, and fibre optic temperature sensor is arranged in the cabinet with contactless mode with mode, the infrared temperature sensor of contact.
Wherein, Fibre optic temperature sensor is the temperature sensor to the harsh and unforgiving environments exploitation of power industry (like power plant, power supply department transformer station, unattended operation transformer station); Adopt the mode of ceramic package, its encapsulating material adopts class of insulation height, high temperature-resistant polymer material.Volume is little, does not influence the normal operation of equipment under test, and easy installation and reliable.
Naked light and electric arc in the main search switch cabinet of photodetector: this detector is with naked light and the such optical signal of electric arc, converts image digital signal into through optical-electrical converter, and applied for machines vision, Digital Image Processing and mode identification technology are confirmed fire.
Central processing unit is realized various signals are monitored, judge, controlled, and is the executing agency that switch cubicle is protected according to preset value.When photodetector detects naked light, the notice central controller, central controller is opened extinguishing device and is put out a fire.When hygrosensor detects the switch in-cabinet temperature and reaches preset value, the notice central processing unit, central processing unit is based on report to the police outwards transmission or start fire extinguishing of setting means.
Above-mentioned cabinet automatic fire extinguishing system adopts following extinguishing method to carry out the identification of fire, may further comprise the steps:
At first be collection to flare up fire:
Because the spectrum 95% of flame light radiation concentrates near infrared band, therefore according to the spectral characteristic of flame, gather the image of this infrared band, thereby get rid of the interference of visible light fire identification.Visible wavelength is generally 380-780nm, and the light of other wave-length coverage is black light, and the flame optical wavelength mainly concentrates on 950-2000nm.
Select the pick-up lens of ccd video camera for use as photoelectric sensor; The black-white CCD video camera not only has stronger spectral response to the visible light of 380-780nm; And the near infrared light of 780-1100nm also had certain spectral response, after adding a slice on the pick-up lens and passing through infrared fileter, make the black-white CCD video camera can only sense the near infrared light that wavelength is 780 ~ 1100nm; In video camera, form black white image; White portion is a flame, and black part is divided into background or other can not get into the jamming light source of camera lens, thus exclusive segment optical jammer source and obtain mainly to comprise the black white image of flame information.
Further because flame has static nature and the behavioral characteristics that is different from optical jammer source; Through Mathematical Modeling and geometrical model with these characteristic quantifications after; Form executable Fire Criterion vector, central processing unit is got rid of optical jammer source and is discerned fire through analyzing the criterion vector.Its behavioral characteristics shows that flame flashes, and flame flashes frequency and mainly concentrates on 7 ~ 12HZ, and great majority rock or fixing interference source height change frequency is not reach 8Hz.Its static nature is that flame is irregular, is angular with respect to other light sources, with circularity as distinguishing rule: C=P
2/ 4 π S, C in the formula---circularity; P---girth; S-area.The circularity of interference source is generally greater than 0.8 less than 1.5, and the circularity of flame is generally greater than 1.8.Central controller as characteristic vector, just can be judged whether breaking out of fire according to characteristic vector with the behavioral characteristics of flame and static nature.Promptly when finding the frequency of flashing to be arranged at 7 ~ 12HZ, circularity promptly determines that it is fire during greater than 1.8 illuminator.
Identification to fire mainly is that the image of detection signal is handled, and referring to Fig. 4, the image of detection signal is handled and comprised the judgement of gathering image, image preliminary treatment, characteristic vector extraction, fire judgement and fire occurrence positions.Reflect the monitoring field condition intuitively through image energy like this, just can discern fire automatically, the record fire at the fire early period of origination.And monitoring range is big, and this is that the conventional fire detector can't be accomplished.
Carry out detailed explanation processing stage of down in the face of each.
1, image preliminary treatment: figure is as shown in Figure 5 for the image pretreatment process, comprises binaryzation, medium filtering, empty filling, target label, correction mark.
1.1 binary conversion treatment is handled according to following binaryzation algorithm:
In the formula:
Binary---binary map pixel gray value;
F---gray-scale map pixel gray value;
T---the binary conversion treatment threshold value makes T=155.
1.2 medium filtering: with target pixel points is that flame pixels and 2n (n is a positive integer) neighborhood territory pixel point gray value carry out descending ordering, gets n+1 value and composes to object pixel, and it is big more that neighborhood is chosen, and filter effect is obvious more, but the also corresponding increase of operand.In binary map because all pixel gray values non-0 promptly 255 can save the process of ordering, if having in object pixel and 2n (n>1) the neighborhood territory pixel gray value less than n+1 individual be 255, the object pixel gray value is changed to 0; Having more than or equal to n+1 is 255, and the object pixel gray value is changed to 255.
1.3 target label:,, need be different gray values with different bright connected component labelings in order to distinguish different bright connected regions owing to possibly have a plurality of bright connected regions in the piece image.Object pixel 8 neighborhood traversal order are as shown in Figure 6.Concrete operations are following:
A, the traversal bianry image, with first object pixel that traverses as sub pixel;
B, with the serve as a mark input parameter of function of sub pixel coordinate, with this point of gray value mark of an appointment;
C, by pixel 8 neighborhood traversal order shown in Figure 6, travel through 8 neighborhoods of seed points one by one, if object pixel stops traversal, invocation flags function self;
After d, intact this connected region of mark, turn back to a step, adopting uses the same method continues other connected region of mark.
1.4 hole fill method: when the inner local luminance of flame was low, some holes can appear in the flame body interior after binary conversion treatment, and these holes are classified as background.Extract the characteristic vector of target area for ease, need to fill these holes, obtain a complete flame main body.
Black region comprises background and hole in the binary map, and wherein background is a maximum black region, and hole is a plurality of little black regions, and background and hole are separated from each other.Utilize region growing algorithm to want earlier to the background area mark, again to the hole zone marker.Because area of flame possibly crossed senior general's background segment and become two black regions; So before mark earlier to view picture figure add that pixel is wide, gray value is 0 frame; Guarantee that first seed points that traverses is a background pixel, and make background can constitute the black connected region of a maximum.
Hole filling algorithm concrete steps based on the background area mark are following:
A, traversing graph picture are that 0 pixel is as sub pixel with a gray value that finds;
B, all pixels of sub pixel place connected region are labeled as Lab [n] (n=0,1,2 through non-recursive algorithm ...);
C, continuing the traversing graph picture, is 0 pixel if traverse a gray value, and n=n+1 gets back to the b step;
After d, view picture figure traveled through completion, background pixel was marked as Lab [0], and hole is labeled as Lab [1], Lab [2], Lab [3] respectively ..., be that the grey scale pixel value of L [0] is changed to 0 with mark value, the grey scale pixel value of non-L [0] is changed to 255, accomplishes hole and fills.
1.5 correction labeling algorithm: because interference is flashed in flame existence on every side or there is hole in inside; Bright area border in 25 two field pictures on the same position can be marked as different gray values; So just can't obtain the characteristic vector of same bright area on 25 two field pictures, so need to revise mark value.
Modification method is to be basic frame with the 0th frame, judge in other frame each boundary marker value whether with the 0th frame same position on the boundary marker value consistent, if inconsistent, with its be revised as with the 0th frame in boundary marker value on the same position.The method can not only be revised the mark value on each border, and can delete the optical jammer source of fast moving or flicker.
2, feature extraction: because the influence of environment, also possibly there is other optical jammer source in detection in the visual field except that flame, thereby influence is to the identification of flame.Though flame and interference source all present the zone of higher brightness in black white image, flame has static nature and the behavioral characteristics of oneself.Static nature is the target signature of extracting through the single-frame images computing, mainly comprises wedge angle, circularity and trapezoidal characteristics.Behavioral characteristics is the target signature of extracting through the multiple image computing, mainly comprise beat, the edge enlarges characteristic.The feature extraction flow process is as shown in Figure 7.
2.1 wedge angle characteristic
Reason owing to flame fringe shake and thermal current rising; Some wedge angles up can appear in flame fringe; The continuous variation of flame wedge angle number is a very significantly performance of flame fringe shake, and almost is constant like the wedge angle number of objects such as candle or electric light.In order to improve the speed of extracting the wedge angle number, adopt the method for based target edge extracting wedge angle number, concrete steps are following:
1) utilizes recurrence method and backtracking method target-marking border, and write down the coordinate sign indicating number of each boundary point;
2) judge according to the coordinate chain code whether each object boundary point is the summit, and can constitute wedge angle with left and right sides neighborhood.
The wedge angle of flame fringe is similar to a triangle, a summit is not only arranged but also certain height is arranged.The summit is exactly the maximum point of flame fringe y coordinate, according to the coordinate chain code that travels through object boundary one by one, relatively its y coordinate whether than about the y coordinate figure of 5,10,15 pixels all big; If satisfy condition, think that then these 31 pixels (15 pixels in the left side, 15 pixels in the right) can constitute a wedge angle.
When object edge generation small variations, all can produce little convexity at random, so will judge also whether the height of wedge angle meets the characteristic of flame wedge angle.After the wedge angle summit was confirmed, as shown in Figure 8, the line between summit, the 15th boundary pixel point of the 15th the boundary pixel point in the left side, summit and the right constituted a triangle, can this leg-of-mutton height be designated as the height of wedge angle.
Triangle length of side a, b, c can obtain through range formula:
Ask leg-of-mutton area according to Heron's formula again:
q=(a+b+c)/2 （4）
Area formula S=c*h/2 substitution formula (formula-3) is got:
When h surpasses preset threshold, can the wedge angle number of this target be added 1, and wedge angle summit grey scale pixel value is changed to 255.Because contiguous back 15 coordinates in this summit possibly also meet the wedge angle condition, but they represent same wedge angle, need jump to 16th boundary pixel point of this summit back and proceed judge when traveling through boundary pixel this moment.
2.2 circularity characteristic: the shape of flame is irregular, and the shape of most of interference source (incandescent lamp, flashlight, candle flame) all is more regular, can describe the regularity of shape through circularity.The computing formula of circularity is following:
In the formula:
C---circularity;
P---girth;
S---area.
* target girth P calculation procedure:
1. adopt recurrence method and backtracking method target-marking border;
2. Freeman 8 direction chain codes coding is carried out on the border, coding rule is as shown in Figure 9;
3. girth is initialized as 0; If chain code is an odd number; Girth adds 1, otherwise girth adds
* target area S computational methods:
Utilize each bright area of onrecurrent region growing method mark, traversal view picture figure asks the area of the corresponding number of pixels of different mark value as each marked region.When the more or single bright area area of bright area number was big, the onrecurrent region growing algorithm was consuming time longer.
The characteristic 2.3 beat: outwardly, it is rambling that flame flashes, and the spectrum signature of flare up fire has distinctive rule in fact.Visually, flame flicking shows as flame and ceaselessly beats, and flame flashes frequency and mainly concentrates on 7 ~ 12Hz, and great majority rock or the fixing source height change frequency that disturbs does not reach 8Hz, can get rid of the lower interference source of height change frequency according to this.
At first extract the height of target in 25 two field pictures, be designated as x [n] (n=0,1,2 ... 24), draw frame number-height oscillogram, then the crest number is counted, the crest number is designated as the target jumping frequency rate according to x [n].
Because fixedly there is slight vibration at the interference source edge; And the height change that slight vibration causes is very little; Generally can not surpass twice target girth; Be that height change generally can not surpass 4 pixels; So have only when crest value during than all big 4 pixels tall of the trough value of both sides; Just carry out frequency counting, the method can obtain the vibration of flame frequency fast.
2.4 trapezoidal characteristics: generally speaking; Because flame base is near comburant; Make the upper width of flame be significantly less than bottom width, its body is similar to a ladder type, generally is to describe this characteristic through the height of C.G. of target and the total height ratio of target; Height of C.G. is the poor of center of gravity ordinate and target lower boundary ordinate minimum of a value, and the target total height is the poor of target coboundary ordinate maximum and lower boundary ordinate minimum of a value.For the regular jamming light source of shape; Its height of C.G. and centre-height are very approaching, and are similar to trapezoidal flame for body, and its height of C.G. is significantly less than centre-height; For shortcut calculation and raising accuracy, this characteristic is described through the distance of center of gravity and central longitudinal coordinate.
The ordinate that regional barycenter is:
In the formula:
Gy---center of gravity ordinate;
S---target area area;
R---target area (being flame region);
Y---the ordinate of each pixel in the target area.
Because image all is based on the border target area is handled, thus the coordinate of target area interior pixels can't directly obtain, so according to the classification and boundary coordinate calculating center of gravity ordinate of boundary point, computational methods are following:
A, utilize target area computational methods reference area S based on the boundary point classification;
B, make
initialization Y=0;
If c, traversal object boundary point classification code are boundary point (x
n, y
n) be the lower boundary point,
If boundary point (x
n, y
n) be the coboundary point,
If boundary point (x
n, y
n) be the horizontal direction summit, Y=Y+y
n(n=0,1,2 wherein ... H-1, the height of h representative image.）
D, the final Y result of Y is updated to (7) formula, obtains the center of gravity ordinate;
E, according to object boundary point ordinate maximin, can calculate the ordinate y of target's center
c:
2.5 the edge enlarges characteristic
At the fire early period of origination, the flame main body increases gradually, and fixedly the main body of interference source is constant basically, and the girth through the target area (calculating identical with above-mentioned target girth P) increases gradually and describes this characteristic.
3, fire is judged: after obtaining the characteristic vector data of flame, need these characteristic synthetics be handled, discriminant parameter is difficult to confirm through manual work.Earlier the sample characteristics vector is input to the BP neutral net; Output layer output valve and ideal value are relatively obtained output error; According to the output error backpropagation, successively calculate the error of each unit of latent layer forward, and with these error correction anterior layer weights.The sample information forward input that so goes round and begins again and the process of error back propagation are performed until output error and reach the acceptable degree, or reach till the predefined study number of times.
Concrete employing Sigmoid type output function, back-propagation algorithm is following:
(1) selected weight coefficient initial value.
(2) repeat following process until restraining (with sample characteristics amount fan-in network successively):
1. calculate each layer unit output o from forward direction
j
2. calculate each elemental error δ of output layer
j
δ
j=(y-o
j) o
j(1-o
j) (formula-11)
3. the reverse transfer error is calculated each latent each elemental error δ of layer
j
4. calculate the correction of each layer weights
△ ω
Ij(t)=α △ ω
Ij(t-1)+η δ
jo
j(formula-13)
5. according to each layer of each layer weights correction correction weights
ω
Ij(t+1)=ω
Ij(t)+△ ω
Ij(t) (formula-14)
Back-propagation algorithm can be revised a latent layer weights, and when adopting gradient method to ask the nonlinear function extreme value, might be absorbed in local minizing point, can not guarantee to converge to global minimum point.
When adopting three layers of feedforward network to carry out the branch time-like, the input layer number is the dimension of characteristic vector; Hide the node layer number and round up for
, wherein p is the input layer number; Output layer node number is a number of categories.In the forward process, the weights initial value chooses usually ± 0.3 interval random number.In the reverse transfer method, step-length η is very big to convergence and optimal value influence, directly sounds out 0.1 ~ 3 usually; Inertia item alpha influences convergence rate, generally between 0.9 ~ 1, selects.
Input layer is done normalization and is handled for the characteristic vector of the infrared light supply of extraction, the output valve of output layer:
(1) out=1 (fire condition)
(2) out=0.5 (precarious position)
(3) out=0 (normal condition)
Perhaps, when using the BP neural network model that fire is differentiated, field condition is differentiated as follows according to output interval value value:
(1) 0.75 ≤out ≤1 fire condition
(2) 0.25 < out < 0.75 precarious positions
(3) 0 ≤out ≤0.25 normal conditions.
Claims (10)
1. cabinet automatic fire extinguishing system; It is characterized in that; Comprise the extinguishing chemical shower nozzle that is installed in the cabinet, the extinguishing chemical shower nozzle has the storage of non-conducting extinguishing agent to be connected through pipeline with storage, and pipeline is provided with the magnetic valve by central controller controls;
Also be provided with photoelectric sensor in the cabinet; Photoelectric sensor comprises pick-up lens and optical-electrical converter; Pick-up lens is provided with infrared fileter, and pick-up lens is taken the formation black white image and converted image digital signal into through optical-electrical converter, and the white portion in the black white image is a flame; Black part is divided into background, and photoelectric sensor outputs to central controller with detection signal;
Central controller is discerned fire according to flame behavioral characteristics and static nature in the middle of the detection signal, and behavioral characteristics characterizes the frequency of flashing of flame, and static nature characterizes the irregularly shaped of flame, differentiates C=P with circularity
2/ 4 π S, wherein C is a circularity, and P is a girth, and S is an area; Flash frequency and during greater than 1.8 illuminator, determine that it is fire at 7～12HZ, circularity when detecting, the central controller controls magnetic valve is opened the release non-conducting extinguishing agent.
2. cabinet automatic fire extinguishing system as claimed in claim 1 is characterized in that, described extinguishing chemical shower nozzle is provided with a plurality of, is connected with storage through the pipeline that is provided with magnetic valve respectively.
3. cabinet automatic fire extinguishing system as claimed in claim 1 is characterized in that, described storage is a pressure storage type extinguishing chemical storage, and its pressure is 1～4Mpa;
Described non-conducting extinguishing agent is heptafluoro-propane or ultra-fine dry powder extinguishing agent.
4. cabinet automatic fire extinguishing system as claimed in claim 1 is characterized in that, also is provided with the hygrosensor that is connected with central controller controls in the described cabinet.
5. cabinet automatic fire extinguishing system as claimed in claim 4; It is characterized in that; Described hygrosensor is fibre optic temperature sensor or infrared temperature sensor, and fibre optic temperature sensor is arranged in the cabinet with contactless mode with mode, the infrared temperature sensor of contact.
6. the fire recognition methods of a cabinet automatic fire extinguishing system is characterized in that, may further comprise the steps:
1) is arranged on the photoelectric sensor photographic images in the cabinet and transmit signal; The pick-up lens of photoelectric sensor is provided with infrared fileter; Pick-up lens is taken the formation black white image and is converted image digital signal into through optical-electrical converter; White portion in the black white image is a flame, and black part is divided into background;
2) central controller receives detection signal; Image digital signal is carried out extracting static nature and the behavioral characteristics that flame is different from optical jammer source after the preliminary treatment; And it is vectorial that it is quantized to form Fire Criterion, gets rid of optical jammer source and discern fire through analyzing the criterion vector;
Described static nature extracts through the single-frame images computing and extracts the target signature that comprises wedge angle, circularity and trapezoidal characteristics:
The wedge angle characteristic: some wedge angles up can appear in flame fringe, and the continuous variation of flame wedge angle number is the performance of flame fringe shake, adopt based target edge extracting wedge angle number;
The circularity characteristic: the shape of flame is irregular, describes the regularity of its shape through circularity, circularity be calculated as C=P
2/ 4 π S, wherein C is a circularity, and P is a girth, and S is an area;
Trapezoidal characteristics: flame base is near comburant, and the upper width that makes flame is described this characteristic less than bottom width through the distance of regional barycenter and central longitudinal coordinate;
Behavioral characteristics be through the multiple image computing extract comprise beat, target signature that the edge enlarges characteristic:
The characteristic of beating: visually, flame flicking shows as flame and ceaselessly beats, and extracts the height of target in 25 two field pictures; Be designated as x [n], n=0,1; 2 ... 24, draw frame number-height oscillogram according to x [n], then the crest number is counted; Have only when crest value during, just carry out frequency counting, the crest number is designated as the target jumping frequency rate than all big 4 pixels tall of the trough value of both sides;
The edge enlarges characteristic: at the fire early period of origination, the flame main body increases gradually, and the girth through the target area increases gradually describes this characteristic;
3) obtain the characteristic vector data of flame after, these characteristic synthetics are handled: earlier static nature and behavioral characteristics vector are input to the BP neutral net; Output layer output valve and ideal value are relatively obtained output error; According to the output error backpropagation, successively calculate the error of each unit of latent layer forward, and with these error correction anterior layer weights; So repeat repeatedly the process of input of sample information forward and error back propagation, be performed until output error and reach the acceptable degree, or reach till the predefined study number of times; Input layer is done normalization and is handled for the characteristic vector of the infrared light supply of extraction, the output valve of output layer, field condition is made the judgement of fire condition, precarious position and normal condition according to output valve.
7. the fire recognition methods of cabinet automatic fire extinguishing system as claimed in claim 6 is characterized in that, described preliminary treatment to image digital signal comprises:
1) binary conversion treatment, handle according to following binaryzation algorithm:
Wherein, Binary is a binary map pixel gray value; F is a gray-scale map pixel gray value; T is the binary conversion treatment threshold value, makes T=155;
2) medium filtering: target pixel points and 2n neighborhood territory pixel point gray value are carried out descending ordering; N is a positive integer; Get n+1 value and compose and give object pixel, if having in object pixel and the 2n neighborhood territory pixel gray value less than n+1 individual be 255, the object pixel gray value is changed to 0; Having more than or equal to n+1 is 255, and the object pixel gray value is changed to 255;
3) target label:, be different gray values with different bright connected component labelings for distinguishing different bright connected regions:
A, the traversal bianry image, with first object pixel that traverses as sub pixel;
B, with the serve as a mark input parameter of function of sub pixel coordinate, with this point of gray value mark of an appointment;
C, travel through 8 neighborhoods of seed points one by one, if object pixel stops traversal, invocation flags function self;
After d, intact this connected region of mark, turn back to a step, adopting uses the same method continues other connected region of mark;
4) hole is filled: when the inner local luminance of flame was low, some holes can appear in the flame body interior after binary conversion treatment, will fill these holes, obtaining a complete flame main body, were filled to based on the hole of background area mark:
A, traversing graph picture are that 0 pixel is as sub pixel with a gray value that finds;
B, all pixels of sub pixel place connected region are labeled as Lab [n] (n=0,1,2 through non-recursive algorithm ...);
C, continuing the traversing graph picture, is 0 pixel if traverse a gray value, and n=n+1 gets back to the b step;
After d, view picture figure traveled through completion, background pixel was marked as Lab [0], and hole is by distinguishing mark, was that the grey scale pixel value of L [0] is changed to 0 with mark value, and the grey scale pixel value of non-L [0] is changed to 255;
5) revise mark: with the 0th frame is basic frame, judge in other frame each boundary marker value whether with the 0th frame same position on the boundary marker value consistent, if inconsistent, with its be revised as with the 0th frame in boundary marker value on the same position.
8. the fire recognition methods of cabinet automatic fire extinguishing system as claimed in claim 6; It is characterized in that; Described wedge angle Feature Extraction is based on object edge and extracts the wedge angle number, utilizes recurrence method and backtracking method target-marking border, and writes down the coordinate sign indicating number of each boundary point; Judge according to the coordinate chain code whether each object boundary point is the summit then, and can constitute wedge angle with left and right sides neighborhood;
The wedge angle of flame fringe is similar to triangle, and the summit is exactly the maximum point of flame fringe y coordinate, according to the coordinate chain code that travels through object boundary one by one, relatively its y coordinate whether than about the y coordinate figure of 5,10,15 pixels all big; If satisfy condition, think that then these pixels constitute a wedge angle;
When object edge generation small variations; To judge also whether the height of wedge angle meets the characteristic of flame wedge angle; After the wedge angle summit is confirmed; Line between summit, the 15th boundary pixel point of the 15th the boundary pixel point in the left side, summit and the right constitutes a triangle, this leg-of-mutton height is designated as the height of wedge angle;
Triangle length of side a, b, c can obtain through range formula:
Ask leg-of-mutton area according to Heron's formula again:
q=(a+b+c)/2 （4）
Area formula S=c*h/2 substitution formula (3) is got:
When h surpasses preset threshold, the wedge angle number of this target is added 1, and wedge angle summit grey scale pixel value is changed to 255; Need jump to 16th boundary pixel point of this summit back and proceed judge when traveling through boundary pixel this moment;
During described circularity feature extraction, girth P is calculated as:
1. adopt recurrence method and backtracking method target-marking border;
2. Freeman 8 direction chain codes coding is carried out on the border;
3. girth is initialized as 0; If chain code is an odd number; Girth adds 1, otherwise girth adds
Area S is calculated as: utilize each bright area of onrecurrent region growing method mark, traversal view picture figure asks the area of the corresponding number of pixels of different mark value as each marked region;
Being calculated as of the distance of the regional barycenter of trapezoidal characteristics and central longitudinal coordinate:
The ordinate of regional barycenter does
Wherein gy is the center of gravity ordinate, and S is the target area area, and R is the target area, and y is the ordinate of each pixel in the target area;
Classification and boundary coordinate according to boundary point calculate the center of gravity ordinate:
A, utilize target area computational methods reference area S based on the boundary point classification;
B, make
initialization Y=0;
If c, traversal object boundary point classification code are boundary point (x
n, y
n) be the lower boundary point,
If boundary point (x
n, y
n) be the coboundary point,
If boundary point (x
n, y
n) be the horizontal direction summit, Y=Y+y
nN=0,1,2 wherein ... H-1, the height of h representative image;
D, the final Y result of Y is updated to (7) formula, obtains the center of gravity ordinate;
E, according to the minimum and maximum value of object boundary point ordinate, calculate the ordinate y of target's center
c:
y
min=Min(y)；y
max=Max(y)；y∈R;
9. the fire recognition methods of cabinet automatic fire extinguishing system as claimed in claim 6 is characterized in that, adopts the backpropagation of Sigmoid type output function to be:
(1) selected weight coefficient initial value;
(2) repeat following process until convergence, with sample characteristics amount fan-in network successively:
1. calculate each layer unit output o from forward direction
j
2. calculate each elemental error δ of output layer
j
δ
j=(y-o
j)o
j(1-o
j) （11）
3. the reverse transfer error is calculated each latent each elemental error δ of layer
j
4. calculate the correction of each layer weights
△ω
ij(t)=α△ω
ij(t-1)+ηδ
jo
j （13）
5. according to each layer of each layer weights correction correction weights
ω
ij(t+1)=ω
ij(t)+△ω
ij(t) （14）。
10. the fire recognition methods of cabinet automatic fire extinguishing system as claimed in claim 6 is characterized in that, when adopting three layers of feedforward network to carry out the branch time-like of fire condition, precarious position and normal condition, the input layer number is the dimension of characteristic vector; Hide the node layer number and round up for
, wherein p is the input layer number; Output layer node number is a number of categories; In the forward process, the weights initial value chooses usually ± 0.3 interval random number; In the reverse transfer method, step-length η sounds out between 0.1～3; Inertia item alpha is selected between 0.9～1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210237440.XA CN102743830B (en) | 2012-07-10 | 2012-07-10 | Automatic electric switch cabinet fire extinguishing system and fire recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210237440.XA CN102743830B (en) | 2012-07-10 | 2012-07-10 | Automatic electric switch cabinet fire extinguishing system and fire recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102743830A true CN102743830A (en) | 2012-10-24 |
| CN102743830B CN102743830B (en) | 2015-01-21 |
Family
ID=47024544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210237440.XA Active CN102743830B (en) | 2012-07-10 | 2012-07-10 | Automatic electric switch cabinet fire extinguishing system and fire recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102743830B (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103423763A (en) * | 2013-07-18 | 2013-12-04 | 武汉九州三维燃烧科技有限公司 | Method for correcting radiation energy signal static deviation |
| CN103845832A (en) * | 2012-11-30 | 2014-06-11 | 珠海格力电器股份有限公司 | Fire extinguishing apparatus and control method thereof, electric box, switch cabinet and electrical equipment |
| CN103861222A (en) * | 2014-02-24 | 2014-06-18 | 钟海平 | Mixed early-warning cabinet fire extinguishing device and early-warning method thereof |
| CN104623831A (en) * | 2014-12-31 | 2015-05-20 | 曙光信息产业(北京)有限公司 | Intelligent fire extinguishing system applied to sever cabinet and devices thereof |
| CN105046868A (en) * | 2015-06-16 | 2015-11-11 | 苏州华启智能科技股份有限公司 | Fire early warning method based on infrared thermal imager in narrow environment |
| CN105261030A (en) * | 2015-11-26 | 2016-01-20 | 四川汇源光通信有限公司 | Method and device for detecting flame from infrared video |
| CN105337180A (en) * | 2015-10-16 | 2016-02-17 | 常熟开关制造有限公司(原常熟开关厂） | Power distribution cabinet and power distribution system |
| CN105788139A (en) * | 2016-03-31 | 2016-07-20 | 嘉兴恒创电力设计研究院有限公司嘉善分公司 | Intelligent fire-extinguishing distribution box |
| CN106231872A (en) * | 2016-08-23 | 2016-12-14 | 江苏明强电气有限公司 | A kind of arc protection is from putting out a fire switch board |
| CN108074368A (en) * | 2016-11-11 | 2018-05-25 | 基德科技公司 | The monitoring based on optical fiber of temperature and/or smoke condition at electronic unit |
| CN110279972A (en) * | 2019-06-28 | 2019-09-27 | 吉林建筑大学 | A kind of electric appliance computer room fire-fighting monitoring device and its control method |
| CN110578987A (en) * | 2019-08-08 | 2019-12-17 | 安徽美博智能电器有限公司 | Cloud-based air conditioner fireproof system and method and storage medium |
| CN110639144A (en) * | 2019-09-20 | 2020-01-03 | 武汉理工大学 | Fire control unmanned ship squirt controlling means based on flame image dynamic identification |
| CN113065421A (en) * | 2021-03-19 | 2021-07-02 | 国网河南省电力公司电力科学研究院 | Multi-source optical fire detection method and system for oil charging equipment |
| CN113222922A (en) * | 2021-04-30 | 2021-08-06 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for detecting light supplement range of infrared flashing lamp |
| US11087606B2 (en) * | 2016-11-11 | 2021-08-10 | Carrier Corporation | High sensitivity fiber optic based detection |
| CN115359617A (en) * | 2022-08-26 | 2022-11-18 | 新创碳谷控股有限公司 | Oxidation furnace flame detection method, computer equipment and storage medium |
| CN117011993A (en) * | 2023-09-28 | 2023-11-07 | 电子科技大学 | Comprehensive pipe rack fire safety early warning method based on image processing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050231354A1 (en) * | 1996-01-23 | 2005-10-20 | Tod Riedel | Remote monitoring |
| CN101315667A (en) * | 2008-07-04 | 2008-12-03 | 南京航空航天大学 | Multi-characteristic synthetic recognition method for outdoor early fire disaster |
| CN101577033A (en) * | 2009-05-26 | 2009-11-11 | 官洪运 | Multiband infrared image-type fire detecting system and fire alarm system thereof |
| CN201603324U (en) * | 2009-12-09 | 2010-10-13 | 上海中发电气（集团）股份有限公司 | Automatic induction and atomization fire-extinguishing system of electric switch cabinet |
-
2012
- 2012-07-10 CN CN201210237440.XA patent/CN102743830B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050231354A1 (en) * | 1996-01-23 | 2005-10-20 | Tod Riedel | Remote monitoring |
| CN101315667A (en) * | 2008-07-04 | 2008-12-03 | 南京航空航天大学 | Multi-characteristic synthetic recognition method for outdoor early fire disaster |
| CN101577033A (en) * | 2009-05-26 | 2009-11-11 | 官洪运 | Multiband infrared image-type fire detecting system and fire alarm system thereof |
| CN201603324U (en) * | 2009-12-09 | 2010-10-13 | 上海中发电气（集团）股份有限公司 | Automatic induction and atomization fire-extinguishing system of electric switch cabinet |
Non-Patent Citations (1)
| Title |
|---|
| 徐仕玲等: "野外早期火灾图像识别方法研究", 《计算机技术与发展》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103845832A (en) * | 2012-11-30 | 2014-06-11 | 珠海格力电器股份有限公司 | Fire extinguishing apparatus and control method thereof, electric box, switch cabinet and electrical equipment |
| CN103423763B (en) * | 2013-07-18 | 2015-12-02 | 武汉九州三维燃烧科技有限公司 | A kind of method revising radiation energy signal static deviation |
| CN103423763A (en) * | 2013-07-18 | 2013-12-04 | 武汉九州三维燃烧科技有限公司 | Method for correcting radiation energy signal static deviation |
| CN103861222A (en) * | 2014-02-24 | 2014-06-18 | 钟海平 | Mixed early-warning cabinet fire extinguishing device and early-warning method thereof |
| CN104623831B (en) * | 2014-12-31 | 2018-08-31 | 曙光节能技术（北京）股份有限公司 | Intelligent fire extinguishing system and device for server cabinet |
| CN104623831A (en) * | 2014-12-31 | 2015-05-20 | 曙光信息产业(北京)有限公司 | Intelligent fire extinguishing system applied to sever cabinet and devices thereof |
| CN105046868A (en) * | 2015-06-16 | 2015-11-11 | 苏州华启智能科技股份有限公司 | Fire early warning method based on infrared thermal imager in narrow environment |
| CN105046868B (en) * | 2015-06-16 | 2019-09-20 | 苏州华启智能科技股份有限公司 | A kind of fire alarm method based on thermal infrared imager in long and narrow environment |
| CN105337180A (en) * | 2015-10-16 | 2016-02-17 | 常熟开关制造有限公司(原常熟开关厂） | Power distribution cabinet and power distribution system |
| CN105261030B (en) * | 2015-11-26 | 2019-01-15 | 四川汇源光通信有限公司 | The method and device of flame is detected from infrared video |
| CN105261030A (en) * | 2015-11-26 | 2016-01-20 | 四川汇源光通信有限公司 | Method and device for detecting flame from infrared video |
| CN105788139A (en) * | 2016-03-31 | 2016-07-20 | 嘉兴恒创电力设计研究院有限公司嘉善分公司 | Intelligent fire-extinguishing distribution box |
| CN106231872A (en) * | 2016-08-23 | 2016-12-14 | 江苏明强电气有限公司 | A kind of arc protection is from putting out a fire switch board |
| CN106231872B (en) * | 2016-08-23 | 2019-04-16 | 江苏明强电气有限公司 | A kind of arc protection is from control cabinet of putting out a fire |
| US10665075B2 (en) | 2016-11-11 | 2020-05-26 | Kidde Technologies, Inc. | Fiber optic based monitoring of temperature and/or smoke conditions at electronic components |
| CN108074368A (en) * | 2016-11-11 | 2018-05-25 | 基德科技公司 | The monitoring based on optical fiber of temperature and/or smoke condition at electronic unit |
| CN108074368B (en) * | 2016-11-11 | 2021-05-07 | 基德科技公司 | Fiber-based monitoring of temperature and/or smoke conditions at electronic components |
| US11087606B2 (en) * | 2016-11-11 | 2021-08-10 | Carrier Corporation | High sensitivity fiber optic based detection |
| CN110279972A (en) * | 2019-06-28 | 2019-09-27 | 吉林建筑大学 | A kind of electric appliance computer room fire-fighting monitoring device and its control method |
| CN110578987A (en) * | 2019-08-08 | 2019-12-17 | 安徽美博智能电器有限公司 | Cloud-based air conditioner fireproof system and method and storage medium |
| CN110578987B (en) * | 2019-08-08 | 2021-02-19 | 安徽美博智能电器有限公司 | Cloud-based air conditioner fireproof system and method and storage medium |
| CN110639144A (en) * | 2019-09-20 | 2020-01-03 | 武汉理工大学 | Fire control unmanned ship squirt controlling means based on flame image dynamic identification |
| CN113065421A (en) * | 2021-03-19 | 2021-07-02 | 国网河南省电力公司电力科学研究院 | Multi-source optical fire detection method and system for oil charging equipment |
| CN113222922A (en) * | 2021-04-30 | 2021-08-06 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for detecting light supplement range of infrared flashing lamp |
| CN113222922B (en) * | 2021-04-30 | 2024-02-27 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for detecting light supplementing range of infrared explosion flash lamp |
| CN115359617A (en) * | 2022-08-26 | 2022-11-18 | 新创碳谷控股有限公司 | Oxidation furnace flame detection method, computer equipment and storage medium |
| CN117011993A (en) * | 2023-09-28 | 2023-11-07 | 电子科技大学 | Comprehensive pipe rack fire safety early warning method based on image processing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102743830B (en) | 2015-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102743830A (en) | Automatic electric switch cabinet fire extinguishing system and fire recognition method | |
| CN111107675B (en) | Cable channel edge Internet of things terminal and method based on ubiquitous power Internet of things | |
| Jazebi et al. | Review of wildfire management techniques—Part I: Causes, prevention, detection, suppression, and data analytics | |
| CN108295407B (en) | Robot cable pipe gallery on-site fire early warning and fire extinguishing method, device and system | |
| CN108898528A (en) | A kind of reasoning by cases method towards hazardous chemical accident emergency aid decision | |
| CN109040278A (en) | Hospital is electrical and dynamical system safe and intelligent manages cloud platform, method and system | |
| CN202876167U (en) | Automatic fire extinguishing system for electrical switch cabinet | |
| CN102509410A (en) | Intelligent fire evacuation escape indicating system and method | |
| KR102114225B1 (en) | Systems and methods for safety management of EES and contacts in electrical transmission and distribution systems | |
| CN109215280A (en) | A kind of electrical fire integrated control system | |
| CN114648853A (en) | Early fire mode identification and grading early warning system of high-voltage switch cabinet | |
| CN106205000A (en) | A kind of portable electric line external force damage prevention monitoring method based on miniradar | |
| KR20170123295A (en) | Forest Fire Monitoring System | |
| CN105488939B (en) | A kind of early warning of fire omnibearing stereo and automatic fire extinguishing system based on intelligent interaction | |
| KR20150061289A (en) | Fire pathway prediction system based USN | |
| Ho et al. | Nighttime fire smoke detection system based on machine vision | |
| CN106781210B (en) | The recognition methods of cigarette in fire based on spatial closure | |
| Shin et al. | Design and implementation of lora-based emergency exit guide system | |
| CN106099689A (en) | A kind of low energy consumption intelligent fire fighting power distribution cabinet | |
| CN206790071U (en) | A kind of novel protected counterpoise grounding electric alarm and protection circuit | |
| KR20150061251A (en) | Fire service system based USN | |
| CN109663258A (en) | A kind of intelligent control fire-fighting system | |
| KR20230107035A (en) | Image recognition-based fire response system and method for power facility | |
| Ivanov et al. | Acoustic Extinguishing of Flames Detected by Deep Neural Networks in Embedded Systems | |
| CN219630478U (en) | Fire extinguishing system of electric vehicle shed |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |