CN110059723A - A kind of robust smog detection method based on integrated depth convolutional neural networks - Google Patents
A kind of robust smog detection method based on integrated depth convolutional neural networks Download PDFInfo
- Publication number
- CN110059723A CN110059723A CN201910206672.0A CN201910206672A CN110059723A CN 110059723 A CN110059723 A CN 110059723A CN 201910206672 A CN201910206672 A CN 201910206672A CN 110059723 A CN110059723 A CN 110059723A
- Authority
- CN
- China
- Prior art keywords
- layer
- pond
- length
- convolution kernel
- full
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
A kind of robust smog detection method based on integrated depth convolutional neural networks not only belongs to field of image recognition, but also belongs to artificial intelligence field.The depth convolutional neural networks of different structure are combined by the method for integrated study and are integrated into classifier by the present invention.The present invention can be to having cigarette smokelessly to detect in factory chimney, torch and other plurality of target scenes.Timely Smoke Detection is carried out, not only can control pollution in industrial circle, it can also be used to the public safety fields such as Forest Fire Alarm.Accuracy rate is improved significantly compared with existing method by exhaust treatment system and other field is applied to based on the robust smog detection method of integrated depth convolutional neural networks, and is avoided existing method and largely adjusted ginseng work.The present invention can generation to exhaust gas and discharge process carry out accurate real-time control, smog is generated and carries out early warning, can not only substantially reduce the discharge of toxic and harmful gas, while can also greatly save manpower.
Description
Technical field
The present invention integrates multiple and different depth convolutional neural networks and establishes the Smoke Detection model for being directed to gray level image, leads to
It crosses using gray level image as input, to thering is non smoke to detect on the image.It based on depth convolutional neural networks and will integrate
The smog detection method of study is applied in exhaust-gas treatment and Smoke Detection, and the generation and discharge process to exhaust gas carry out accurate
Real-time control, can not only substantially reduce the discharge of toxic and harmful gas, while also reduce energy consumption, save human resources,
Improve production efficiency.Smog detection method based on depth convolutional neural networks and integrated study had both belonged to image recognition neck
Domain, and belong to artificial intelligence field.
Background technique
In recent years, China has been devoted to protection environment, and energy-saving and emission-reduction reduce atmosphere pollution.New edition " ambient air quality
Standard " and respectively for the discharge standard of the industries such as thermoelectricity, steel, cement, chemical industry and non-electrical coal-burning boiler be promote it is each
The most strong policy of enterprise implement smoke treated engineering.But since the industries such as traditional thermoelectricity, chemical industry account in China's economy
There is biggish specific gravity, the discharge amount of exhaust gas is still very big, and lacks the side detected to the exhaust gas of the discharges such as torch, chimney
Method leads to air pollution problems inherent still and annoyings many enterprises.
The method of traditional smoke detection relies primarily on artificial observation or sensor.But since human resources are limited, cost compared with
Height, the method based on artificial observation cannot fast and effeciently monitor smog for a long time.On the other hand, due to the influence of environmental change,
Smoke sensor device based on smoke particle sampling or relative humidity sampling is also likely to will appear serious time lag, while also not
Detection zone can be completely covered.In general, existing smog detection method is difficult to meet demand.
In recent years, significant progress is achieved using the technology that convolutional neural networks carry out image recognition, especially with
The raising of modern computer operational capability can be efficiently extracted using depth convolutional neural networks by the study to great amount of samples
Clarification of objective is to realize accurate image recognition.But since the setting of the design parameter of neural network does not standardize, so
Leading to the neural network under different structure, parameter often has biggish performance gap, simultaneously because the diversity of permutation and combination,
It is difficult to determine best structure after all being tested the network of all categories.Based on problem above, the invention proposes integrated
The neural network of multiple and different structures improves the accuracy of algorithm, also avoids to both realize the diversity of classifier
Repetition test is also difficult to the problem of determining optimum network structure.
Summary of the invention
The present invention integrates multiple and different depth convolutional neural networks and establishes the Smoke Detection model for being directed to gray level image, leads to
Gray level image to detect target is crossed as input, to thering is non smoke to detect on the image.It is detected by this method,
The problem of not only being promoted obviously in accuracy rate compared with existing method, also avoid tuning parameter repeatedly, passes through to realize
The smog image of input carries out accurate, detection in real time to smog.Essence is carried out for the discharge process to combustion process and exhaust gas
True ground real-time control creates condition；
Present invention employs the following technical solution and realize step:
1. a kind of robust smog detection method based on integrated depth convolutional neural networks:
There is no smog to be detected for depositing in target scene, using the gray level image of target scene as input；
Characterized by comprising the following steps:
(1) multiple sub- depth convolutional neural networks are designed and train, the input of these networks is gray level image, this is a little
Neural network needs to embody the diversity of structure, and this patent devises three kinds of basic neural network structures: DN11, DN8,
DN5, and 10 different neural networks are respectively generated with them, it is total to have obtained the sub-neural network of 30 different structures.
The structure of DN11 are as follows: the first layer of network is convolutional layer to the 4th layer, and the dimension that convolution kernel uses is U, convolution kernel
Number be V, the step-length of convolution is 1, using ReLU activation primitive, to the characteristic pattern of input using full 0 filling, each layer into
Row batch normalization；Layer 5 is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；Layer 6 to the 8th layer be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；9th layer is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；The
Ten layers to the 13rd layer are convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is used
ReLU activation primitive fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；14th layer is pond layer,
Pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；15th layer and the 16th layer is full connection
Layer, there is 2048 neurons, and dropout probability is 0.5；Finally by softmax function output category result, there are cigarette or nothing
Cigarette.
The structure of DN8 are as follows: the first layer of network to third layer be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；4th layer is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；Layer 5 and layer 6 are convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；Layer 7 is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；The
Eight layers to the tenth layer are convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is used
ReLU activation primitive fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；Eleventh floor is pond layer,
Pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Floor 12 and the 13rd layer are full connection
Layer, there is 2048 neurons, and dropout probability is 0.5；Finally by softmax function output category result, there are cigarette or nothing
Cigarette.
The structure of DN5 are as follows: the first layer and the second layer of network be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；Third layer is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；4th layer is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V,
The step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, carries out batch normalizing in each layer
Change；Layer 5 is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Layer 6 and
Layer 7 is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is activated using ReLU
Function fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；8th layer is pond layer, using 2 dimensions
Pond range and maximum pond, horizontal and vertical step-length is 2；9th layer and the tenth layer is full articulamentum, there is 2048 minds
Through member, dropout probability is 0.5；Finally by softmax function output category result, there is cigarette or smokeless.
In the neural network structure on three of the above basis, the value by the way that different U and V is arranged can be obtained by different
Sub-neural network.This patent is 3,5,7,9,11, V 32,16 by setting U, by the value permutation and combination life of different U and V
At each 10 different sub-networks of these three network structures of DN11, DN8, DN5, to obtain amounting to 30 sub-neural networks.
(2) 30 sub-neural networks of generation are subjected to integrated and beta pruning, specific step are as follows:
1. selecting the network that precision is best in 30 sub-networks first to be put into integrated classifier.Lead to from 30 networks again
It crosses iteration and finds out next second son for being added to integrated classifier and integrated classifier capable of being made to be optimal precision in verifying collection
Network repeats above step q times, each sub-network can at most be repeated selection n times, we set 2 for the value of N here,
To obtain the integrated classifier containing q sub-network.
2. all not selected sub-networks are established into set, then repeatedly band successively by integrated classifier sub-network and
Not selected sub-network swaps, if integrated classifier precision improvement, just swaps the two, it is on the contrary then not
Become.
3. change band (1), (2) step 30 times are repeated, the value that can achieve the q of full accuracy is found out by q value from 5-35 value,
And using the highest integrated classifier of this precision as final integrated classifier.
Integrated classifier is thus obtained.
Creativeness of the invention is mainly reflected in:
(1) present invention can reflect the feature of image for the neural network of different structure from different perspectives, so integrated
The sub-neural network of multiple and different structures is to reach higher accuracy；
(2) present invention for existing neural network algorithm need it is a large amount of adjust join work aiming at the problem that, by integrate it is multiple not
The neural network of same parameter avoids adjusts ginseng repeatedly, reduces workload；
Detailed description of the invention
Fig. 1 is structure chart of the invention
Specific embodiment
The present invention integrates multiple and different depth convolutional neural networks and establishes the Smoke Detection model for being directed to gray level image, leads to
Gray level image to detect target is crossed as input, to thering is non smoke to detect on the image.It is detected by this method,
The problem of not only being promoted obviously in accuracy rate compared with existing method, also avoid tuning parameter repeatedly, passes through to realize
The smog image of input carries out accurate, detection in real time to smog.Essence is carried out for the discharge process to combustion process and exhaust gas
True ground real-time control creates condition；
Present invention employs the following technical solution and realize step:
1. designing and training multiple sub- depth convolutional neural networks, the input of these networks is gray level image, this is a little
Neural network needs to embody the diversity of structure, and this patent devises three kinds of basic neural network structures: DN11, DN8, DN5,
And 10 different neural networks are respectively generated with them, it is total to have obtained the sub-neural network of 30 different structures.
The structure of DN11 are as follows: the first layer of network is convolutional layer to the 4th layer, and the dimension that convolution kernel uses is U, convolution kernel
Number be V, the step-length of convolution is 1, using ReLU activation primitive, to the characteristic pattern of input using full 0 filling, each layer into
Row batch normalization；Layer 5 is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；Layer 6 to the 8th layer be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；9th layer is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；The
Ten layers to the 13rd layer are convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is used
ReLU activation primitive fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；14th layer is pond layer,
Pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；15th layer and the 16th layer is full connection
Layer, there is 2048 neurons, and dropout probability is 0.5；Finally by softmax function output category result, there are cigarette or nothing
Cigarette.
The structure of DN8 are as follows: the first layer of network to third layer be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；4th layer is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；Layer 5 and layer 6 are convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；Layer 7 is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；The
Eight layers to the tenth layer are convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is used
ReLU activation primitive fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；Eleventh floor is pond layer,
Pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Floor 12 and the 13rd layer are full connection
Layer, there is 2048 neurons, and dropout probability is 0.5；Finally by softmax function output category result, there are cigarette or nothing
Cigarette.
The structure of DN5 are as follows: the first layer and the second layer of network be convolutional layer, the dimension that convolution kernel uses for U, convolution kernel
Number is V, and the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, is carried out at each layer
Criticize normalization；Third layer is pond layer, and pond range and maximum pond using 3 dimensions fill out the characteristic pattern of input using full 0
It fills, horizontal and vertical step-length is 2；4th layer is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V,
The step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, carries out batch normalizing in each layer
Change；Layer 5 is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Layer 6 and
Layer 7 is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is activated using ReLU
Function fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；8th layer is pond layer, using 2 dimensions
Pond range and maximum pond, horizontal and vertical step-length is 2；9th layer and the tenth layer is full articulamentum, there is 2048 minds
Through member, dropout probability is 0.5；Finally by softmax function output category result, there is cigarette or smokeless.
In the neural network structure on three of the above basis, the value by the way that different U and V is arranged can be obtained by different
Sub-neural network.This patent is 3,5,7,9,11, V 32,16 by setting U, by the value permutation and combination life of different U and V
At each 10 different sub-networks of these three network structures of DN11, DN8, DN5, thus obtain amounting to 30 sub-neural networks, and
9016 width smog images and 8363 width non smoke images have been used to train above all of neural network.
2. 30 sub-neural networks of generation are carried out integrated and beta pruning, specific step are as follows:
(1) network that precision is best in 30 sub-networks is selected first to be put into integrated classifier.Again from 30 networks
Finding out next integrated classifier that is added to by iteration can make integrated classifier be optimal second of precision in verifying collection
Sub-network repeats above step q times, each sub-network can at most be repeated selection n times, we set the value of N to here
2, to obtain the integrated classifier containing q sub-network, verifying collection has smog image and 8511 non smoke figures by 8804
As composition.
(2) all not selected sub-networks are established into set, then repeatedly band successively by the sub-network in integrated classifier
Swapped with not selected sub-network, if integrated classifier precision improvement, just swaps the two, it is on the contrary then
It is constant.
(3) by q value from 5-35 value, change band (1), (2) step 30 times is repeated, the value that can achieve the q of full accuracy is found out
It is 15, and using the highest integrated classifier of this precision as final integrated classifier.
Thus obtained integrated classifier, through testing our integrated classifier, by 1240 Zhang Youyan images and
In the test set of 1648 smokeless image constructions, accuracy rate has reached 98.71%.
Claims (1)
1. a kind of robust smog detection method based on integrated depth convolutional neural networks, which comprises the following steps:
Step 1: designing the sub- convolutional neural networks of multiple depth and training；
Step 2: establishing the learner of integrated neural network and being trimmed, negative sense sub-neural network is removed；
In the first step:
Multiple sub- depth convolutional neural networks are designed and train, the input of these networks is gray level image, this little nerve net
Network needs to embody the diversity of structure, and this patent devises three kinds of basic neural network structures: DN11、DN8、DN5, and use it
Respectively generate 10 different neural networks, it is total to have obtained the sub-neural network of 30 different structures；
DN11Structure are as follows: the first layer of network to the 4th layer be convolutional layer, the dimension that convolution kernel uses is U, the number of convolution kernel
For V, the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, criticize in each layer and return
One changes；Layer 5 is pond layer, and pond range and maximum pond using 3 dimensions fill the characteristic pattern of input using full 0, horizontal
It is 2 to the step-length with longitudinal direction；Layer 6 is convolutional layer to the 8th layer, and for U, the number of convolution kernel is the dimension that convolution kernel uses
V, the step-length of convolution are 1, using ReLU activation primitive, are filled to the characteristic pattern of input using full 0, carry out batch normalizing in each layer
Change；9th layer is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Tenth layer is arrived
13rd layer is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is swashed using ReLU
Function living fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；14th layer is pond layer, using 2
The pond range of dimension and maximum pond, horizontal and vertical step-length is 2；15th layer and the 16th layer is full articulamentum, is had
2048 neurons, dropout probability are 0.5；Finally by softmax function output category result, there is cigarette or smokeless；
DN8Structure are as follows: the first layer of network to third layer be convolutional layer, the dimension that convolution kernel uses is U, the number of convolution kernel
For V, the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, criticize in each layer and return
One changes；4th layer is pond layer, and pond range and maximum pond using 3 dimensions fill the characteristic pattern of input using full 0, horizontal
It is 2 to the step-length with longitudinal direction；Layer 5 and layer 6 are convolutional layer, and for U, the number of convolution kernel is the dimension that convolution kernel uses
V, the step-length of convolution are 1, using ReLU activation primitive, are filled to the characteristic pattern of input using full 0, carry out batch normalizing in each layer
Change；Layer 7 is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；8th layer is arrived
Tenth layer is convolutional layer, and for the dimension that convolution kernel uses for U, the number of convolution kernel is V, and the step-length of convolution is 1, is activated using ReLU
Function fills the characteristic pattern of input using full 0, carries out batch normalization in each layer；Eleventh floor is pond layer, using 2 dimensions
Pond range and maximum pond, horizontal and vertical step-length is 2；Floor 12 and the 13rd layer are full articulamentum, are had
2048 neurons, dropout probability are 0.5；Finally by softmax function output category result, there is cigarette or smokeless；
DN5Structure are as follows: the first layer and the second layer of network be convolutional layer, the dimension that convolution kernel uses is U, the number of convolution kernel
For V, the step-length of convolution is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, criticize in each layer and return
One changes；Third layer is pond layer, and pond range and maximum pond using 3 dimensions fill the characteristic pattern of input using full 0, horizontal
It is 2 to the step-length with longitudinal direction；4th layer is convolutional layer, and the dimension that convolution kernel uses is U, and the number of convolution kernel is V, convolution
Step-length is 1, using ReLU activation primitive, is filled to the characteristic pattern of input using full 0, carries out batch normalization in each layer；5th
Layer is pond layer, and pond range and maximum pond using 2 dimensions, horizontal and vertical step-length is 2；Layer 6 and layer 7 are
Convolutional layer, the dimension that convolution kernel uses is U, and the number of convolution kernel is V, and the step-length of convolution is 1, right using ReLU activation primitive
The characteristic pattern of input is filled using full 0, carries out batch normalization in each layer；8th layer is pond layer, using the pond range of 2 dimensions
With maximum pond, horizontal and vertical step-length is 2；9th layer and the tenth layer is full articulamentum, there is 2048 neurons,
Dropout probability is 0.5；Finally by softmax function output category result, there is cigarette or smokeless；
In the neural network structure on three of the above basis, setting U is 3,5,7,9,11, V 32,16, by different U and V
Value permutation and combination generates DN11、DN8、DN5Each 10 different sub-networks of these three network structures, to obtain 30 total
Sub-neural network；
30 sub-neural networks of generation are subjected to integrated and beta pruning, specific step are as follows:
(1) network that precision is best in 30 sub-networks is selected first to be put into integrated classifier；Pass through from 30 networks again
Iteration finds out next second subnet for being added to integrated classifier and integrated classifier capable of being made to be optimal precision in verifying collection
Network repeats above step q times, each sub-network can at most be repeated selection n times, we set 2 for the value of N here, from
And obtain the integrated classifier containing q sub-network；
(2) all not selected sub-networks are established into set, then repeatedly band is not successively by the sub-network in integrated classifier and
Selected sub-network swaps, if integrated classifier precision improvement, just swaps the two, it is on the contrary then not
Become；
(3) by q value from 5-35 value, change band (1), (2) step 30 times is repeated, the value that can achieve the q of full accuracy is found out, and
Using the highest integrated classifier of this precision as final integrated classifier；
Integrated classifier is thus obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910206672.0A CN110059723B (en) | 2019-03-19 | 2019-03-19 | Robust smoke detection method based on integrated deep convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910206672.0A CN110059723B (en) | 2019-03-19 | 2019-03-19 | Robust smoke detection method based on integrated deep convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110059723A true CN110059723A (en) | 2019-07-26 |
| CN110059723B CN110059723B (en) | 2021-01-05 |
Family
ID=67317210
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910206672.0A Active CN110059723B (en) | 2019-03-19 | 2019-03-19 | Robust smoke detection method based on integrated deep convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059723B (en) |
Cited By (4)
| 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 |
| CN112418005A (en) * | 2020-11-06 | 2021-02-26 | 北京工业大学 | Smoke multi-classification identification method based on backward radiation attention pyramid network |
| CN112801187A (en) * | 2021-01-29 | 2021-05-14 | 广东省科学院智能制造研究所 | Hyperspectral data analysis method and system based on attention mechanism and ensemble learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012041333A1 (en) * | 2010-09-30 | 2012-04-05 | Visiopharm A/S | Automated imaging, detection and grading of objects in cytological samples |
| CN106228150A (en) * | 2016-08-05 | 2016-12-14 | 南京工程学院 | Smog detection method based on video image |
| 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 |
| JP2018101416A (en) * | 2016-12-21 | 2018-06-28 | ホーチキ株式会社 | Fire monitoring system |
| CN109086803A (en) * | 2018-07-11 | 2018-12-25 | 南京邮电大学 | A kind of haze visibility detection system and method based on deep learning and the personalized factor |
| CN109271906A (en) * | 2018-09-03 | 2019-01-25 | 五邑大学 | A kind of smog detection method and its device based on depth convolutional neural networks |
| CN109376695A (en) * | 2018-11-26 | 2019-02-22 | 北京工业大学 | A kind of smog detection method based on depth hybrid neural networks |
-
2019
- 2019-03-19 CN CN201910206672.0A patent/CN110059723B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012041333A1 (en) * | 2010-09-30 | 2012-04-05 | Visiopharm A/S | Automated imaging, detection and grading of objects in cytological samples |
| CN106228150A (en) * | 2016-08-05 | 2016-12-14 | 南京工程学院 | Smog detection method based on video image |
| JP2018101416A (en) * | 2016-12-21 | 2018-06-28 | ホーチキ株式会社 | Fire monitoring system |
| 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 |
| CN109086803A (en) * | 2018-07-11 | 2018-12-25 | 南京邮电大学 | A kind of haze visibility detection system and method based on deep learning and the personalized factor |
| CN109271906A (en) * | 2018-09-03 | 2019-01-25 | 五邑大学 | A kind of smog detection method and its device based on depth convolutional neural networks |
| CN109376695A (en) * | 2018-11-26 | 2019-02-22 | 北京工业大学 | A kind of smog detection method based on depth hybrid neural networks |
Non-Patent Citations (2)
| Title |
|---|
| YU CHUN-YU等: "Video fire smoke detection using motion and color features", 《FIRE TECHNOLOGY》 * |
| 陈俊周等: "基于级联卷积神经网络的视频动态烟雾检测", 《电子科技大学学报》 * |
Cited By (6)
| 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 |
| CN112418005A (en) * | 2020-11-06 | 2021-02-26 | 北京工业大学 | Smoke multi-classification identification method based on backward radiation attention pyramid network |
| CN112418005B (en) * | 2020-11-06 | 2024-05-28 | 北京工业大学 | Smoke multi-classification identification method based on reverse radiation attention pyramid network |
| CN112801187A (en) * | 2021-01-29 | 2021-05-14 | 广东省科学院智能制造研究所 | Hyperspectral data analysis method and system based on attention mechanism and ensemble learning |
| CN112801187B (en) * | 2021-01-29 | 2023-01-31 | 广东省科学院智能制造研究所 | Hyperspectral data analysis method and system based on attention mechanism and ensemble learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110059723B (en) | 2021-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059723A (en) | A kind of robust smog detection method based on integrated depth convolutional neural networks | |
| CN110909483B (en) | Point source atmospheric pollutant emission list verification method based on gridding data | |
| CN108064047B (en) | Water quality sensor network optimization deployment method based on particle swarm | |
| US20230194755A1 (en) | Data-driven rapid traceability method for air pollutants in small-scale regionals | |
| CN110907066B (en) | Wind turbine generator gearbox bearing temperature state monitoring method based on deep learning model | |
| CN111814111A (en) | Industrial park atmospheric pollutant tracing method | |
| CN106873359B (en) | Wind power noise evaluation method based on cluster analysis and neural network | |
| CN104200089B (en) | Method for measuring air pollutant emission amount of coal burned at bungalows | |
| CN106208046B (en) | A kind of tidal current energy generating field unit layout method considering power generation settings cost | |
| CN107239613A (en) | A kind of intelligent source class recognition methods based on online data and Factor Analysis Model | |
| CN109376695B (en) | Smoke detection method based on deep hybrid neural network | |
| CN112287294B (en) | Space-time bidirectional soil water content interpolation method based on deep learning | |
| CN105116730B (en) | Hydrogen-fuel engine electronic spark advance angle and optimizing system and its optimization method based on Particle Group Fuzzy Neural Network | |
| CN109087277A (en) | A kind of air fine particles PM2.5 measurement method based on characteristics of image and integrated neural network | |
| CN115656446B (en) | Air quality detection system and method based on Internet of things | |
| CN110738357A (en) | Method, device and system for predicting dust collection amount of coal yard and storage medium | |
| Zhou et al. | Combined estimation of fire perimeters and fuel adjustment factors in FARSITE for forecasting wildland fire propagation | |
| CN109409666B (en) | Environmental impact assessment method based on atmospheric diffusion model and linear programming | |
| CN109766905A (en) | Target cluster dividing method based on Self-Organizing Feature Maps | |
| CN113657023A (en) | Near-surface ozone concentration inversion method based on combination of machine learning and deep learning | |
| Rangel et al. | An assessment of dispersing pollutants from the pre-harvest burning of sugarcane in rural areas in the northeast of Brazil | |
| CN112241800B (en) | Method for calculating VOCs pollutant emission amount of coke oven | |
| CN114519124A (en) | Joint defense and joint control treatment method for atmospheric environmental pollution | |
| CN109059870B (en) | Boiler atmosphere pollutant emission monitoring system and inspection method based on unmanned aerial vehicle aerial image | |
| CN110967458A (en) | Method for detecting emission range of atmospheric pollutants in industrial area |
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 |