CN109598891A - A kind of method and system for realizing Smoke Detection using deep learning disaggregated model - Google Patents
A kind of method and system for realizing Smoke Detection using deep learning disaggregated model Download PDFInfo
- Publication number
- CN109598891A CN109598891A CN201811578770.9A CN201811578770A CN109598891A CN 109598891 A CN109598891 A CN 109598891A CN 201811578770 A CN201811578770 A CN 201811578770A CN 109598891 A CN109598891 A CN 109598891A
- Authority
- CN
- China
- Prior art keywords
- layer
- image
- smog
- gone
- module
- 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
- 239000000779 smoke Substances 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 238000001514 detection method Methods 0.000 title claims abstract description 24
- 238000006243 chemical reaction Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 33
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 6
- 235000019504 cigarettes Nutrition 0.000 claims description 3
- 239000003595 mist Substances 0.000 claims 2
- 230000000694 effects Effects 0.000 abstract description 16
- 238000005259 measurement Methods 0.000 abstract description 7
- 238000012795 verification Methods 0.000 abstract description 7
- 238000011897 real-time detection Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000009738 saturating Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
- G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of methods for realizing Smoke Detection using deep learning disaggregated model, it include: that frame smog image to be gone is obtained from video flowing, smog image is treated using gauss hybrid models to be handled, to obtain the moving region of the smog image to be gone, image is handled using dark defogging algorithm, to obtain smokeless iconic model, obtain the error image between smog image to be gone and smokeless iconic model, binary conversion treatment is carried out to error image, to obtain doubtful smoke region, obtain the intersection area between moving region and doubtful smoke region, intersection area is inputted in trained deep learning disaggregated model, to obtain final smog recognition result, and smoke region is marked in smog image to be gone according to the smog recognition result.The deep learning disaggregated model of present invention lightweight reaches higher accuracy rate and verification and measurement ratio, reduces false detection rate, and the effect of real-time detection may be implemented.
Description
Technical field
The invention belongs to technical field of image processing, are realized more particularly, to a kind of using deep learning disaggregated model
The method and system of Smoke Detection.
Background technique
Fire is one of highest disaster of probability of happening in natural calamity and social disaster, is constituted to human lives and life
It seriously threatens.The early period that fire occurs is usually associated with the generation of a large amount of smog, as can timely and accurately smog is detected,
There is far reaching significance to fire alarm and fighting.
Nowadays the video smoke detection method mainly used is become based on color characteristic, edge detection, LBP operator, small echo
The manual features extracting method changed etc., but there is some can not ignore in these detection methods: the first, due to smog
Shape and color-variable, the extracted feature generalization ability of these methods is poor, lower so as to cause its accuracy rate and verification and measurement ratio, accidentally
Inspection rate is higher, and application scenarios are limited;The second, these methods are generally required using entire image as input, computationally intensive, real-time
It is poor;Third, common deep learning disaggregated model more preferable recognition effect in order to obtain, generally use more complicated model knot
Structure, more so as to cause model parameter amount, algorithm is computationally intensive.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, invention introduces deep learning algorithms, by character representation
Optimize with Classifier combination, provides a kind of method and system for realizing Smoke Detection using deep learning disaggregated model, instructing
When practicing disaggregated model, the score cluster loss function proposed in the present invention is added, it is intended that with the deep learning of lightweight
Disaggregated model reaches higher accuracy rate and verification and measurement ratio, reduces false detection rate, and the effect of real-time detection may be implemented.
To achieve the above object, according to one aspect of the present invention, it provides a kind of real using deep learning disaggregated model
The method of existing Smoke Detection, comprising the following steps:
(1) frame smog image I to be gone is obtained from video flowing;
(2) smog image to be gone I obtained in step (1) is handled using gauss hybrid models, with obtain this to
Remove the moving region Region1 of smog image I;
(3) image I is handled using dark defogging algorithm, to obtain smokeless iconic model J;
(4) difference between the smokeless iconic model J that the obtained smog image I to be gone of obtaining step (1) and step (3) obtain
It is worth image P;
(5) binary conversion treatment is carried out to the error image P that step (4) obtains, to obtain doubtful smoke region Region2;
(6) doubtful smoke region obtained in moving region Region1 and step (4) obtained in obtaining step (2)
Intersection area between Region2;
(7) intersection area obtained in step (6) is inputted in trained deep learning disaggregated model, it is final to obtain
Smog recognition result, and smoke region is marked in smog image I to be gone according to the smog recognition result.
Preferably, step (3) is specifically to use following formula:
Wherein x indicates that any one of image I pixel, A indicate overall atmosphere light intensity, and t (x) indicates that image I's is saturating
Penetrate rate.
Preferably, transmissivityWherein c indicates the RGB color in image I
Channel, Ω (x) indicate the square area centered on pixel x in the c of RGB color channel, and y indicates square area Ω
(x) pixel in, Ic(y) image in RGB channel c, A are indicatedcIndicate the overall atmosphere light intensity in RGB color channel.
Preferably, the binary conversion treatment process in step (5) is specifically to use following formula:
Wherein T indicates binarization threshold, and value range is between 25 to 35 preferably 30.
Preferably, deep learning disaggregated model is generated by following procedure:
(a) it is concentrated from smoke data and obtains multiple smog images and non-smog image;
(b) randomly selected m image in all images obtained in step (a) is instructed using convolutional neural networks
Practice, to obtain discrimination of the convolutional neural networks on smog test set, wherein the value of m is 200 to 500;
(c) it for all images obtained in step (a), constantly repeats the above steps (b), until convolutional neural networks exist
Until discrimination on smog test set reaches maximum value, to obtain trained convolutional neural networks.
Preferably, convolutional neural networks use ten layers of structure, and first layer is input layer, and input is the pixel of 32*32*3
Matrix, the second layer are the first convolutional layers, receive the picture element matrix of the 32*32*3 from input layer, and wherein convolution kernel is 3*3*
32, which is filled using full 0, step-length 1, this layer of output matrix size is 32*32*32, and third layer is the first pond layer, Chi Hua
Window size is 2*2, and long and wide step-length is 2, this layer of output matrix is 16*16*32, and the 4th layer is the second convolutional layer, volume
Product core is filled having a size of 3*3*64, step-length 1, the layer using full 0, and the matrix of output is 16*16*64, and layer 5 is the second pond
Change layer, pond window size is 2*2, and the matrix size of step-length 2, output is 8*8*64, and layer 6 is third convolutional layer, convolution
Core is filled having a size of 3*3*128, step-length 1, the layer using full 0, and the matrix of output is 8*8*128, and layer 7 is third pond
Layer, pond window size are 2*2, and step-length 2, the matrix size of output is 4*4*128, and the 8th layer is full articulamentum, output
Node number is 512, does a non-linear conversion for the output feature to third pond layer, to obtain feature vector, the 9th
Layer is linear weighted function layer, and output node number is 2, does linear weighted function for the output to full articulamentum, is obtained with obtaining classification
Divide vector Si, the tenth layer is loss function layer.
Preferably, loss function used in loss function layer includes cross entropy loss function and score cluster loss letter
Number.
Preferably, score cluster loss function LSCIt indicates are as follows:
Wherein SiIndicating category score vector, dimension is identical with classification number, per the one-dimensional score for representing a classification,
SCyiIndicate yiThe score center vector of classification, dimension and category score vector SiIt is identical, it is initialized as when training for the first time complete
0 vector is updated during each training.
A kind of realize that Smoke Detection is using deep learning disaggregated model it is another aspect of this invention to provide that providing
System, comprising:
First module, for obtaining frame smog image I to be gone from video flowing;
Second module, for being handled using gauss hybrid models the smog image I to be gone that the first module obtains, with
Obtain the moving region Region1 of the smog image I to be gone;
Third module, for being handled using dark defogging algorithm image I, to obtain smokeless iconic model J;
4th module, for obtaining the smog image I to be gone that the first module obtains and the smokeless image that third module obtains
Error image P between model J;
5th module, the error image P for obtaining to the 4th module carries out binary conversion treatment, to obtain doubtful smog area
Domain Region2;
6th module, for obtaining the moving region Region1 that the second module obtains and the doubtful cigarette that the 4th module obtains
Intersection area between the Region2 of fog-zone domain;
7th module, the intersection area for obtaining the 6th module input in trained deep learning disaggregated model,
To obtain final smog recognition result, and smoke region is marked in smog image I to be gone according to the smog recognition result.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
1, since present invention employs step (2) to arrive step (6), preliminary screening has been done to smoke region, only needs to doubt
Convolutional neural networks are inputted like smoke region, rather than to entire image, to solve in existing video smoke detection method
It is existing computationally intensive, the poor technical problem of real-time, while the above process can also exclude a part of jamming target, reduce
False detection rate.
2, since present invention employs step (5), the smog characteristics of image of robust out can be learnt automatically, therefore can solve
It is poor that feature generalization ability is manually extracted present in certainly existing video smoke detection method, cause its accuracy rate and verification and measurement ratio compared with
Technical problem low, false detection rate is higher, application scenarios are limited.
3, it joined score cluster loss function, in convolutional neural networks model due to the present invention so as to not change
In the case where varying model structure, model accuracy rate and verification and measurement ratio are further increased, and reduce false detection rate.
4, score cluster loss function joined in convolutional neural networks model due to the present invention, can simplify model, use
Relatively simple network structure can reach the recognition effect of complex network, to reduce the parameter amount of model, go forward side by side one
Step improves efficiency of algorithm, realizes the effect of real-time detection.
5, in step (5), with the raising of convolutional neural networks model recognition effect, convolutional neural networks mould is increased
Spacing between the class for the feature vector that full articulamentum extracts in type, so as to learn better smog characteristics of image out.
Detailed description of the invention
Fig. 1 is the flow chart for the method that the present invention realizes Smoke Detection using deep learning disaggregated model.
The smog image to be gone obtained in the step of Fig. 2 shows the method for the present invention (1);
Fig. 3 shows error image obtained in the step of the method for the present invention (4);
Fig. 4 shows doubtful smoke region obtained in the step of the method for the present invention (5);
Fig. 5 shows the smoke region marked in the step of the method for the present invention (7);
Fig. 6 is the schematic block diagram of convolutional neural networks used in the method for the present invention;
Fig. 7 (a) and Fig. 7 (b) is using the score distributed effect figure before and after score cluster loss function of the present invention respectively;
Fig. 8 (a) and Fig. 8 (b) is imitated using the feature distribution extracted before and after score cluster loss function of the present invention respectively
Fruit figure.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
As shown in Figure 1, the present invention using deep learning disaggregated model realize Smoke Detection method the following steps are included:
(1) frame smog image I to be gone is obtained from video flowing, as shown in Figure 2;
(2) smog image to be gone I obtained in step (1) is handled using gauss hybrid models, with obtain this to
Remove the moving region Region1 of smog image I;
(3) dark defogging algorithm (Single Image Haze Removal Using Dark Channel is utilized
Prior) image I is handled, to obtain smokeless iconic model J;
Specifically, this step is specifically to use formula below:
Wherein x indicates that any one pixel inside image I, A indicate overall atmosphere light intensity, and t (x) indicates image I's
Transmissivity, and haveWherein c indicates the RGB color channel in image I, Ω (x)
Indicate the square area centered on pixel x in the c of RGB color channel, y indicates the picture in square area Ω (x)
Vegetarian refreshments, Ic(y) image in RGB channel c, A are indicatedcIndicate the overall atmosphere light intensity in RGB color channel.
(4) difference between the smokeless iconic model J that the obtained smog image I to be gone of obtaining step (1) and step (3) obtain
It is worth image P (as shown in Figure 3);
Specifically, obtaining error image in this step is using following formula:
P (x)=| I (x)-J (x) |
(5) binary conversion treatment is carried out to the error image P that step (4) obtains, to obtain doubtful smoke region Region2
(as shown in Figure 4);
Binary conversion treatment is specifically to use following formula:
In above formula, T indicates binarization threshold, and value range is between 25 to 35 preferably 30.
(6) doubtful smoke region obtained in moving region Region1 and step (4) obtained in obtaining step (2)
Intersection area between Region2,
(7) intersection area obtained in step (6) is inputted in trained deep learning disaggregated model, it is final to obtain
Smog recognition result, and smoke region is marked in smog image I to be gone according to the smog recognition result.
After this step process, the smoke region effect finally marked is as shown in the box of Fig. 5.
Deep learning disaggregated model in the present invention is generated by following procedure: from smoke data collection, (it is to pass through
The mode of crawler on network from getting) in obtain multiple smog images and non-smog image, utilize convolutional neural networks pair
Image is trained randomly selected m a (wherein the value of m is 200 to 500), and constantly repeats this mistake in this multiple images
Journey, until convolutional neural networks are until the discrimination on smog test set reaches best, to obtain trained convolution mind
Through network.
As shown in fig. 6, convolutional neural networks of the invention include input layer, three convolutional layers, three pond layers, one it is complete
Articulamentum, a linear weighted function layer and a loss function layer.
First layer is input layer, inputs the picture element matrix for 32*32*3.
The second layer is the first convolutional layer, receives the picture element matrix of the 32*32*3 from input layer, and wherein convolution kernel is 3*
3*32, the layer are filled using full 0, step-length 1, this layer of output matrix size is 32*32*32;
Third layer is the first pond layer, and pond window size is 2*2, and long and wide step-length is 2, this layer of output matrix is
16*16*32;
4th layer is the second convolutional layer, and convolution kernel is filled having a size of 3*3*64, step-length 1, the layer using full 0, output
Matrix is 16*16*64;
Layer 5 is the second pond layer, and pond window size is 2*2, and step-length 2, the matrix size of output is 8*8*64;
Layer 6 is third convolutional layer, and convolution kernel is filled having a size of 3*3*128, step-length 1, the layer using full 0, output
Matrix is 8*8**128;
Layer 7 is third pond layer, and pond window size is 2*2, and step-length 2, the matrix size of output is 4*4*128;
8th layer is full articulamentum, and output node number is 512, is done once for the output feature to third pond layer
Non-linear conversion, to obtain feature vector;
9th layer is linear weighted function layer, and output node number is 2, does linear weighted function for the output to full articulamentum,
To obtain category score vector Si;
Tenth layer is loss function layer.
Wherein, loss function used in the loss function layer of above-mentioned convolutional neural networks include cross entropy loss function and
Score cluster loss function.
Specifically, score cluster loss function L of the inventionSCIt indicates are as follows:
Wherein SiIndicating category score vector, (its dimension is identical with classification number, represents obtaining for a classification per one-dimensional
Point), SCyiIndicate yiScore center vector (its dimension and the category score vector S of classificationiIt is identical, it is initialized when training for the first time
For full 0 vector, it is updated during each training).
Test result and compare
Here actual effect of the invention is illustrated by the test on smoke data collection.
(1) score distributed effect compares
As shown in Fig. 7 (a) and (b), wherein " x " represents smog image, label is [1,0], should be located at two-dimensional coordinate bottom right
Angle, "○" represent non-smog sample, and label is [0,1], should be located at the two-dimensional coordinate upper left corner.Fig. 7 (a) is not plus score cluster is damaged
The score distribution of function is lost, the non-smog image in part is classified mistake, causes false detection rate higher, and Fig. 7 (b) is clustered plus score
Score distribution after loss function, it can be seen that the non-smog image of mistake is classified in Fig. 7 (a) in score cluster loss letter
Under the action of number, cluster arrives the upper left corner, to keep its classification correct.
(2) feature distribution Contrast on effect
The promotion of model recognition effect passes through the feature that gradient propagates backward to the 8th layer (i.e. full articulamentum) output simultaneously
Vector increases spacing between class so as to improve the effect for the feature extracted.Fig. 8 (a) is not add score cluster loss function
Feature distribution, it can be seen that spacing is smaller between class, and many images are classified mistake, and Fig. 8 (b) is that added score cluster loss letter
Feature distribution after number, spacing is larger between class, and only a few part is classified mistake.
(3) comparison of accuracy rate, verification and measurement ratio, false detection rate, parameter amount and classical convolution disaggregated model
As shown in table 1 below, it illustrates the method for the present invention and existing AlexNet, VGG16 classical taxonomy model and
DNCNN disaggregated model disclosed in IEEE paper is in accuracy rate, the comparison of verification and measurement ratio, false detection rate, parameter amount these aspect of performance:
Table 1
By upper table 1 it can be seen that
(a) in the case where unmodified model, three indexs of disaggregated model of the invention are not all than adding score cluster damage
The effect of mistake improves very much;
(b) in the comparison with AlexNet and VGG16 classical taxonomy model, disaggregated model of the invention is in accuracy rate
And the effect of false detection rate is all best;
(c) compared with DNCNN disaggregated model result, three indexs of disaggregated model of the invention will be better than its effect;
(d) disaggregated model parameter amount of the invention is least, and parameter amount means that the raising of efficiency of algorithm less, in phase
In the same time, it can handle more video datas.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (9)
1. a kind of method for realizing Smoke Detection using deep learning disaggregated model, which comprises the following steps:
(1) frame smog image I to be gone is obtained from video flowing;
(2) smog image to be gone I obtained in step (1) is handled using gauss hybrid models, to obtain the cigarette to be gone
The moving region Region1 of mist image I;
(3) image I is handled using dark defogging algorithm, to obtain smokeless iconic model J;
(4) differential chart between the smokeless iconic model J that the obtained smog image I to be gone of obtaining step (1) and step (3) obtain
As P;
(5) binary conversion treatment is carried out to the error image P that step (4) obtains, to obtain doubtful smoke region Region2;
(6) doubtful smoke region Region2 obtained in moving region Region1 and step (4) obtained in obtaining step (2)
Between intersection area;
(7) intersection area obtained in step (6) is inputted in trained deep learning disaggregated model, to obtain final cigarette
Mist recognition result, and smoke region is marked in smog image I to be gone according to the smog recognition result.
2. the method according to claim 1, wherein step (3) is specifically to use following formula:
Wherein x indicates that any one of image I pixel, A indicate overall atmosphere light intensity, and t (x) indicates the transmissivity of image I.
3. according to the method described in claim 2, it is characterized in that, transmissivity
Wherein c indicate image I in RGB color channel, Ω (x) indicate in the c of RGB color channel centered on pixel x just
Square region, y indicate the pixel in square area Ω (x), Ic(y) image in RGB channel c, A are indicatedcIndicate RGB face
Overall atmosphere light intensity in chrominance channel.
4. the method according to claim 1, which is characterized in that the binary conversion treatment mistake in step (5)
Journey is specifically to use following formula:
Wherein T indicates binarization threshold, and value range is between 25 to 35 preferably 30.
5. method as claimed in any of claims 1 to 4, which is characterized in that deep learning disaggregated model is to pass through
What following procedure generated:
(a) it is concentrated from smoke data and obtains multiple smog images and non-smog image;
(b) randomly selected m image in all images obtained in step (a) is trained using convolutional neural networks,
To obtain discrimination of the convolutional neural networks on smog test set, wherein the value of m is 200 to 500;
(c) it for all images obtained in step (a), constantly repeats the above steps (b), until convolutional neural networks are in smog
Until discrimination on test set reaches maximum value, to obtain trained convolutional neural networks.
6. according to the method described in claim 5, it is characterized in that,
Convolutional neural networks use ten layers of structure;
First layer is input layer, and input is the picture element matrix of 32*32*3.
The second layer is the first convolutional layer, receives the picture element matrix of the 32*32*3 from input layer, and wherein convolution kernel is 3*3*
32, which is filled using full 0, step-length 1, this layer of output matrix size is 32*32*32;
Third layer is the first pond layer, and pond window size is 2*2, and long and wide step-length is 2, this layer of output matrix is 16*
16*32;
4th layer is the second convolutional layer, and convolution kernel is filled having a size of 3*3*64, step-length 1, the layer using full 0, the matrix of output
For 16*16*64;
Layer 5 is the second pond layer, and pond window size is 2*2, and step-length 2, the matrix size of output is 8*8*64;
Layer 6 is third convolutional layer, and convolution kernel is filled having a size of 3*3*128, step-length 1, the layer using full 0, the matrix of output
For 8*8*128;
Layer 7 is third pond layer, and pond window size is 2*2, and step-length 2, the matrix size of output is 4*4*128;
8th layer is full articulamentum, and output node number is 512, is done for the output feature to third pond layer primary non-thread
Property conversion, to obtain feature vector;
9th layer is linear weighted function layer, and output node number is 2, does linear weighted function for the output to full articulamentum, with
To category score vector Si;
Tenth layer is loss function layer.
7. according to the method described in claim 6, it is characterized in that, loss function used in loss function layer includes cross entropy
Loss function and score cluster loss function.
8. the method according to the description of claim 7 is characterized in that score cluster loss function LSCIt indicates are as follows:
Wherein SiIndicating category score vector, dimension is identical with classification number, per the one-dimensional score for representing a classification, SCyi
Indicate yiThe score center vector of classification, dimension and category score vector SiIt is identical, for the first time training when be initialized as full 0 to
Amount is updated during each training.
9. a kind of system for realizing Smoke Detection using deep learning disaggregated model characterized by comprising
First module, for obtaining frame smog image I to be gone from video flowing;
Second module, for being handled using gauss hybrid models the smog image I to be gone that the first module obtains, to obtain
The moving region Region1 of the smog image I to be gone;
Third module, for being handled using dark defogging algorithm image I, to obtain smokeless iconic model J;
4th module, for obtaining the smog image I to be gone that the first module obtains and the smokeless iconic model J that third module obtains
Between error image P;
5th module, the error image P for obtaining to the 4th module carries out binary conversion treatment, to obtain doubtful smoke region
Region2;
6th module, for obtaining the moving region Region1 that the second module obtains and the doubtful smog area that the 4th module obtains
Intersection area between the Region2 of domain;
7th module, the intersection area for obtaining the 6th module input in trained deep learning disaggregated model, with
Smoke region is marked in smog image I to be gone to final smog recognition result, and according to the smog recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811578770.9A CN109598891B (en) | 2018-12-24 | 2018-12-24 | A kind of method and system for realizing Smoke Detection using deep learning disaggregated model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811578770.9A CN109598891B (en) | 2018-12-24 | 2018-12-24 | A kind of method and system for realizing Smoke Detection using deep learning disaggregated model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109598891A true CN109598891A (en) | 2019-04-09 |
CN109598891B CN109598891B (en) | 2019-10-08 |
Family
ID=65963150
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811578770.9A Active CN109598891B (en) | 2018-12-24 | 2018-12-24 | A kind of method and system for realizing Smoke Detection using deep learning disaggregated model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109598891B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
CN111008612A (en) * | 2019-12-24 | 2020-04-14 | 标旗(武汉)信息技术有限公司 | Production frequency statistical method, system and storage medium |
CN111046827A (en) * | 2019-12-20 | 2020-04-21 | 哈尔滨理工大学 | Video smoke detection method based on convolutional neural network |
CN111103477A (en) * | 2019-10-31 | 2020-05-05 | 武汉旗云高科信息技术有限公司 | Electrical load monitoring method and system based on 3D convolutional neural network |
CN111126136A (en) * | 2019-11-18 | 2020-05-08 | 上海交通大学 | Smoke concentration quantification method based on image recognition |
CN111274896A (en) * | 2020-01-15 | 2020-06-12 | 深圳市守行智能科技有限公司 | Smoke and fire identification algorithm |
CN111462451A (en) * | 2019-11-01 | 2020-07-28 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
CN111563488A (en) * | 2020-07-14 | 2020-08-21 | 成都市映潮科技股份有限公司 | Video subject content identification method, system and storage medium |
CN111783900A (en) * | 2020-07-10 | 2020-10-16 | 宁波方太厨具有限公司 | Training method of oil smoke concentration detection model and control method of range hood gear |
CN112198170A (en) * | 2020-09-29 | 2021-01-08 | 合肥公共安全技术研究院 | Detection method for identifying water drops in three-dimensional detection of outer surface of seamless steel pipe |
CN113869164A (en) * | 2021-09-18 | 2021-12-31 | 的卢技术有限公司 | Smoke detection method based on deep learning |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101084719B1 (en) * | 2010-06-25 | 2011-11-22 | (주)퓨처아이스 | Intelligent smoke detection system using image processing and computational intelligence |
US8472699B2 (en) * | 2006-11-22 | 2013-06-25 | Board Of Trustees Of The Leland Stanford Junior University | Arrangement and method for three-dimensional depth image construction |
CN106250845A (en) * | 2016-07-28 | 2016-12-21 | 北京智芯原动科技有限公司 | Flame detecting method based on convolutional neural networks and device |
CN106934404A (en) * | 2017-03-10 | 2017-07-07 | 深圳市瀚晖威视科技有限公司 | A kind of image flame identifying system based on CNN convolutional neural networks |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
KR101869442B1 (en) * | 2017-11-22 | 2018-06-20 | 공주대학교 산학협력단 | Fire detecting apparatus and the method thereof |
CN108229458A (en) * | 2017-12-22 | 2018-06-29 | 湖南源信光电科技股份有限公司 | A kind of intelligent flame recognition methods based on motion detection and multi-feature extraction |
CN108428324A (en) * | 2018-04-28 | 2018-08-21 | 温州大学激光与光电智能制造研究院 | The detection device of smog in a kind of fire scenario based on convolutional network |
CN108647601A (en) * | 2018-04-28 | 2018-10-12 | 温州大学激光与光电智能制造研究院 | The detection method of smog in a kind of fire scenario based on convolutional network |
-
2018
- 2018-12-24 CN CN201811578770.9A patent/CN109598891B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8472699B2 (en) * | 2006-11-22 | 2013-06-25 | Board Of Trustees Of The Leland Stanford Junior University | Arrangement and method for three-dimensional depth image construction |
KR101084719B1 (en) * | 2010-06-25 | 2011-11-22 | (주)퓨처아이스 | Intelligent smoke detection system using image processing and computational intelligence |
CN106250845A (en) * | 2016-07-28 | 2016-12-21 | 北京智芯原动科技有限公司 | Flame detecting method based on convolutional neural networks and device |
CN106934404A (en) * | 2017-03-10 | 2017-07-07 | 深圳市瀚晖威视科技有限公司 | A kind of image flame identifying system based on CNN convolutional neural networks |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
KR101869442B1 (en) * | 2017-11-22 | 2018-06-20 | 공주대학교 산학협력단 | Fire detecting apparatus and the method thereof |
CN108229458A (en) * | 2017-12-22 | 2018-06-29 | 湖南源信光电科技股份有限公司 | A kind of intelligent flame recognition methods based on motion detection and multi-feature extraction |
CN108428324A (en) * | 2018-04-28 | 2018-08-21 | 温州大学激光与光电智能制造研究院 | The detection device of smog in a kind of fire scenario based on convolutional network |
CN108647601A (en) * | 2018-04-28 | 2018-10-12 | 温州大学激光与光电智能制造研究院 | The detection method of smog in a kind of fire scenario based on convolutional network |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111103477A (en) * | 2019-10-31 | 2020-05-05 | 武汉旗云高科信息技术有限公司 | Electrical load monitoring method and system based on 3D convolutional neural network |
CN111462451A (en) * | 2019-11-01 | 2020-07-28 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
CN111462451B (en) * | 2019-11-01 | 2022-04-26 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
CN111126136A (en) * | 2019-11-18 | 2020-05-08 | 上海交通大学 | Smoke concentration quantification method based on image recognition |
CN111126136B (en) * | 2019-11-18 | 2023-04-21 | 上海交通大学 | Smoke concentration quantification method based on image recognition |
CN111046827A (en) * | 2019-12-20 | 2020-04-21 | 哈尔滨理工大学 | Video smoke detection method based on convolutional neural network |
CN111008612A (en) * | 2019-12-24 | 2020-04-14 | 标旗(武汉)信息技术有限公司 | Production frequency statistical method, system and storage medium |
CN111274896A (en) * | 2020-01-15 | 2020-06-12 | 深圳市守行智能科技有限公司 | Smoke and fire identification algorithm |
CN111274896B (en) * | 2020-01-15 | 2023-09-26 | 深圳市守行智能科技有限公司 | Smoke and fire recognition algorithm |
CN111783900A (en) * | 2020-07-10 | 2020-10-16 | 宁波方太厨具有限公司 | Training method of oil smoke concentration detection model and control method of range hood gear |
CN111563488A (en) * | 2020-07-14 | 2020-08-21 | 成都市映潮科技股份有限公司 | Video subject content identification method, system and storage medium |
CN112198170A (en) * | 2020-09-29 | 2021-01-08 | 合肥公共安全技术研究院 | Detection method for identifying water drops in three-dimensional detection of outer surface of seamless steel pipe |
CN112198170B (en) * | 2020-09-29 | 2023-09-29 | 合肥公共安全技术研究院 | Detection method for identifying water drops in three-dimensional detection of outer surface of seamless steel tube |
CN113869164A (en) * | 2021-09-18 | 2021-12-31 | 的卢技术有限公司 | Smoke detection method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN109598891B (en) | 2019-10-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109598891B (en) | A kind of method and system for realizing Smoke Detection using deep learning disaggregated model | |
CN110348376B (en) | Pedestrian real-time detection method based on neural network | |
CN108399362A (en) | A kind of rapid pedestrian detection method and device | |
CN112906485B (en) | Visual impairment person auxiliary obstacle perception method based on improved YOLO model | |
CN109493303A (en) | A kind of image defogging method based on generation confrontation network | |
CN108388896A (en) | A kind of licence plate recognition method based on dynamic time sequence convolutional neural networks | |
Gunawan et al. | Sign language recognition using modified convolutional neural network model | |
CN109543760A (en) | Confrontation sample testing method based on image filters algorithm | |
CN110334660A (en) | A kind of forest fire monitoring method based on machine vision under the conditions of greasy weather | |
CN111488805B (en) | Video behavior recognition method based on salient feature extraction | |
CN112686276A (en) | Flame detection method based on improved RetinaNet network | |
CN106339984A (en) | Distributed image super-resolution method based on K-means driven convolutional neural network | |
CN115690730A (en) | High-speed rail contact net foreign matter detection method and system based on single classification and abnormal generation | |
CN117765480B (en) | Method and system for early warning migration of wild animals along road | |
CN115240259A (en) | Face detection method and face detection system based on YOLO deep network in classroom environment | |
EP3798897A1 (en) | Methods for artificial neural networks | |
CN104616034B (en) | A kind of smog detection method | |
CN114155551A (en) | Improved pedestrian detection method and device based on YOLOv3 under complex environment | |
CN113221667A (en) | Face and mask attribute classification method and system based on deep learning | |
CN113902044B (en) | Image target extraction method based on lightweight YOLOV3 | |
CN110991243A (en) | Straw combustion identification method based on combination of color channel HSV and convolutional neural network | |
CN116385368A (en) | Photovoltaic cell defect detection data set augmentation method based on generation countermeasure network | |
CN112257800B (en) | Visual identification method based on deep convolutional neural network model-regeneration network | |
CN109461162A (en) | The method of Target Segmentation in image | |
CN111260570B (en) | Binarization background noise simulation method for posts based on cyclic consistency confrontation network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |