CN108921039A - The forest fire detection method of depth convolution model based on more size convolution kernels - Google Patents
The forest fire detection method of depth convolution model based on more size convolution kernels Download PDFInfo
- Publication number
- CN108921039A CN108921039A CN201810584066.8A CN201810584066A CN108921039A CN 108921039 A CN108921039 A CN 108921039A CN 201810584066 A CN201810584066 A CN 201810584066A CN 108921039 A CN108921039 A CN 108921039A
- Authority
- CN
- China
- Prior art keywords
- layer
- model
- convolution
- size
- forest fire
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 23
- 230000003287 optical effect Effects 0.000 claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000000034 method Methods 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 230000000644 propagated effect Effects 0.000 claims description 2
- 230000004048 modification Effects 0.000 claims 1
- 238000012986 modification Methods 0.000 claims 1
- 210000004218 nerve net Anatomy 0.000 claims 1
- 230000005540 biological transmission Effects 0.000 abstract description 10
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 description 8
- 230000033001 locomotion Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 102000008297 Nuclear Matrix-Associated Proteins Human genes 0.000 description 1
- 108010035916 Nuclear Matrix-Associated Proteins Proteins 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 210000000299 nuclear matrix Anatomy 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000012905 visible particle Substances 0.000 description 1
- 239000002023 wood Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- 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
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/005—Fire alarms; Alarms responsive to explosion for forest fires, e.g. detecting fires spread over a large or outdoors area
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Emergency Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Fire-Detection Mechanisms (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of forest fire detection methods of depth convolution model based on more size convolution kernels, it is related to deep learning video identification technology field, including video image information acquisition, data wireless transmission, optical flow field are formed, convolutional neural networks model foundation and prediction of result in deep learning.The depth convolutional neural networks model that the present invention passes through the different size convolution kernels of training, it improves existing algorithm and combines existing hardware condition, accelerate the speed of forest fire judgement, improve the precision of prediction of entire model, effectively help to solve the problems, such as existing forest fire discovery not in time, economic loss it is more serious, there is certain practical value.
Description
Technical field
The invention belongs to video network technology identification technology and artificial intelligence fields, and in particular to a kind of based on more to establishing
The depth convolutional neural networks model of size convolution kernel uses optical flow field to carry out as monitor video of the input of model to forest
The method that fire condition judges and predicts fire trend.
Background technique
It mainly include leading portion video 1. video identification technology is widely used since effect is good, data acquisition is convenient
The video analysis processing three phases of acquisition and transmission, the intermediate video detection and rear end of information.What video identification technology was constituted
Video stream network is extensive, unattended, efficient accurately information collection mode, by fixed or shift position in forest
The floating of visible particles in the imaging of light stream in air in network high-definition camera automatic collection forest, abnormal light and air
Situation etc. is to the advantageous information for judging fire.Collected information is transmitted through the network to monitoring center and forms accurate reality
When monitor.
2. artificial intelligence is all more popular research direction of current every field, wherein containing a variety of subjects and calculation
Method.Wherein obtaining the artificial neural network of good result is the basic principle based on neural network in biology, is understanding and is taking out
After human brain structure and environmental stimuli response mechanism, using network topology as theoretical basis, the nervous system of human brain is simulated to multiple
A kind of mathematical model of the treatment mechanism of miscellaneous information, inherence can be obtained from existing data by not needing any priori formula
Rule, the ability with certain cluster and prediction.But in recent years, simple artificial neural network has been unable to meet the mankind's
Demand, the gradually research by scholar for image recognition etc. of convolutional neural networks model.
3. forest fire is one of the disaster for needing to attract great attention to the entire ecosystem, no matter its cause is
It is natural or artificial, suffer from high risk.Due to the particularity of forest, once fire occurs, fire spreading speed is fast
And be not easy to be extinguished, so the prevention of forest fire just seems particularly significant.Therefore, the prediction of forest fire can be accomplished to prevent
Suffer from possible trouble, relevant departments can quickly take corresponding reply decision.Currently, China carries out the measure master of forest fire monitoring
To be used is satellite remote sensing and the method manually gone on patrol, but the costly of satellite remote sensing, positioning are inaccurate；Work as use
The method manually gone on patrol needs to expend a large amount of manpower and material resources if forest occupied area is very wide, and timeliness is poor.
Summary of the invention
To solve the problems, such as that forest fire finds that speed is slow, treatment effect is poor, the present invention provides a kind of based on more sizes
The forest fire detection method of the depth convolutional neural networks model of convolution kernel, can use manpower and material resources sparingly to a certain extent.
The present invention solve above-mentioned technical problem the technical solution adopted is that：
A kind of forest fire detection method of the depth convolution model based on more size convolution kernels, it is characterised in that including such as
Lower step：
【1】Acquire the video image of forest；
【2】Using optical flow method, a width light stream field picture is generated per adjacent two frame video image, and by the optical flow field figure of generation
Image set is divided into training set and test set in proportion；
【3】The depth convolutional neural networks model for establishing more size convolution kernels, by step【2】Obtained training set is as mind
Input through network model, is trained model, by step【2】Obtained test set is used for the precision of prediction of testing model,
Continue to optimize model；
【4】The light stream field picture for generating region to be predicted, detects Fires Occurred using the model after optimization.
Preferably, the depth convolutional neural networks model of more size convolution kernels include input layer, several convolutional layers,
Several pond layers, full articulamentum and output layer；The input layer is for receiving light stream field picture；The convolutional layer is used for will be defeated
Enter the received light stream field picture of layer and feature extraction is carried out by convolution kernel；Each convolutional layer connects a pond layer and carries out down
Sampling, the characteristic image that convolutional layer obtains all are used as the input of pond layer to carry out Further Feature Extraction, and pond layer is to characteristic pattern
As the feature progress aggregate statistics of different location, to obtain the lower image of dimension；Upper one layer of pond layer and next layer of convolution
Layer connection, is repeated feature extraction, and to the last one layer of pond layer exports final characteristic image to full articulamentum；Full articulamentum
All features are connected, and is conveyed to output layer and classifies；Output layer finally obtains classification results using classifier.
The present invention can efficiently, accurately judge whether that fire occurs using detection method of the neural network in conjunction with optical flow field
Calamity simultaneously analyzes the scale of fire, type and predicts its development trend, relevant department can carry out in advance the corresponding precautionary measures or
Fire controls fire behavior when occurring as early as possible.Image Acquisition network based on video can effectively acquire the high-definition image of forest, will
Data are real-time transmitted to command centre and carry out data processing again, and the system cost is low, is easily achieved and with the high spy of accuracy
Point.Therefore the fire hazard monitoring, fire scheme forming and fire that this method can highly desirable solve a wide range of forest occur
Implementation issue afterwards.
Compared with prior art, significant advantage of the present invention is：(1) it is detected automatically using deep learning instead of existing
Artificial detection, it is time saving and energy saving, improve detection efficiency；(2) the light stream field picture that input layer is obtained using optical flow method, reduces
The time of neural network judgement movement or floating body position, training and predetermined speed of entire model are improved, model is made to exist
Moving object distinguishes that aspect precision is higher.(3) it is improved for basic convolutional neural networks model, use is various sizes of
Convolution kernel makes entire model reach iteration termination condition faster, improvement falls into Local Minimum problem, it is slow etc. to solve convergence rate
The FAQs of model.(4) present invention incorporates video information acquisition, wireless network data transmission and real-time management response systems
System can find fire location, prediction fire tendency more quickly and take corresponding fire suppression measures rapidly.
Detailed description of the invention
Fig. 1 is the system structure signal of the forest fire detection method of the depth convolution model based on more size convolution kernels
Figure.
Fig. 2 is the training set of model and the establishment process figure of test set.
Fig. 3 is the model training flow chart of the depth convolutional neural networks of more size convolution kernels.
Fig. 4 is the flow chart of the forest fire detection method of the depth convolution model based on more size convolution kernels.
Specific embodiment
The explanation of specific embodiment is carried out to the present invention with reference to the accompanying drawing.
Fig. 1 is the system structure signal of the forest fire detection method of the depth convolution model based on more size convolution kernels
Figure mainly includes video surveillance network and forest fire detection model two parts.The former is used to shoot the real-time status in forest
And by transmission of video to control centre；The latter according to collected data judges whether that fire occurs and becoming for fire occurs
Gesture.
Monitoring network is mainly realized by the network high-definition camera for being mounted on suitable position in forest, captured by camera
Content need to cover full wafer forest, main to shoot dynamic substance or object in forest, such as：Smog flame etc., and by collected view
Frequency saves, is transferred to control centre.The transmission mode that video image taken by network high-definition camera passes through wireless bridge
It is transferred to the crucial coordinator node of each network, is then transmitted to control centre, forest fire detection model is established in control
In the host of the heart.
The core of this method is to form the optical flow field of image and establishes the depth convolution mind based on more size convolution kernels
Through network modelling two parts.Specifically, being exactly that captured picture is passed through wireless network by the high-definition camera of wood land
It is transmitted to control centre, the latter will form optical flow field and be supplied to prediction model together with other data after image procossing, by mould
The prediction of type judges, obtains the probability of time point fire generation, then indicates the region pole when the probability is greater than a certain threshold value
Forest fire may occur greatly, relevant department is needed to take measures on customs clearance as early as possible.
System bottom is made of the camera and sensor being placed in forest monitoring region, for acquiring the base of forest
Notebook data, and use the transmission of wireless bridge progress data.Wireless bridge is wireless radio-frequency and traditional limited bridge skill
The product that art combines.Due to its adaptability is good, transmission range farther out, the transmission mode of this kind of wireless bridge can be fine
Adapt to forest in monitor video and other data transmission services.Since transmission range may be farther out so in using having
Wireless bridge after station transmits, and data are transferred to relay station by collection point first, then are transferred to control centre by relay station, side
Just relevant administrative staff's processing.
The variation of image is expressed in light stream, since it contains target motion information, can be used to determine with observed person
The motion conditions of target.The present embodiment forms optical flow field according to the forest image of acquisition, due to the aerial expansion of flame smog
The diffusion different from other substances is dissipated, optical flow field can record the motion change of object in the picture just.Use optical flow method
Form the basic ideas of optical flow field：Observe video image it can be found that temporally adjacent image there is regular hour phases
Closing property and position correlation, the motion information of object can be calculated using the correlation.Specifically, if (u, v) is image
(x, y, u, v), then be known as light stream point by the light stream of point (x, y), and the collection of all light stream points is collectively referred to as optical flow field.Therefore optical flow field can
To regard two-dimensional instantaneous velocity field as, to state the motion conditions of object within a certain period of time.
The reference of optical flow field can clearly be judged in adjacent two images with the presence or absence of the object of movement, optical flow field
Image has increased considerably the accuracy for forest fire prediction after the classification of convolutional neural networks model.
The present invention is based on the forest fire detection methods of the depth convolution model of more size convolution kernels, and the method includes such as
Lower step：
Step 1：Equipment is installed.Suitable position installation network high-definition camera is chosen in forest, passes through wireless network
Relay station is transferred data to, then control centre is transmitted to by relay station.
Step 2：Video image optical flow field is formed.The video that network high-definition camera acquires in real time saves as time company by frame
Continuous picture carries out the pretreatment operations such as gray processing to picture.The Horn-Schunck algorithm based on gradient is reused to calculate
The light stream of adjacent image, detailed algorithm are as follows：
It is E (x, y, t) in the brightness of t moment image assuming that a pixel (x, y) on image, while with u (x, y) and v
(x, y) indicates the light stream, and component in the horizontal and vertical directions uses following formula subrepresentation respectively：
U=dx/dt
V=dy/dt
The image after multiframe, the brightness of the corresponding point after interval of time Δ t, i.e., are being corresponded in the present embodiment
For E, (x+ Δ x, y+ Δ y, t+ Δ t), wherein Δ x, Δ y are the changing value within the Δ t period in both direction.When Δ t very little
And level off to 0 when, it is believed that the brightness of the point is constant, so there is following formula：
E (x, y, t)=E (x+ Δ x, y+ Δ y, t+ Δ t)
If illustrating that the point was moved or changed within this time when the brightness of the point changes, moving
The brightness put afterwards is unfolded by Taylor's formula, can be obtained：
Second order infinitesimal ε wherein can be ignored then to be had due to Δ t → 0：
Therefore above formula is basic optical flow constraint equation, if enabling：
Indicate that pixel is along x in grayscale image, y, the gradient on the direction t, therefore can be with
The optical flow constraint equation of Horn-Schunck optical flow computation method is rewritten as：
Exu+Eyv+Et=0
According to the light stream field picture of adjacent two field pictures in the available video of above-mentioned algorithm.In proportion by resulting image
(the present embodiment presses 7:3 ratio) it is divided into training set and test set, input of the training set as the model of subsequent foundation optimizes mould
Parameters in type；Whether precision of prediction of the test set for judgment models meets the requirements.
Step 3：Design the depth convolutional neural networks of more size convolution kernels.
The depth convolutional neural networks model of more size convolution kernels is successively by input layer, several convolutional layers, several ponds
Layer, full articulamentum and output layer are constituted.Wherein input layer is formed by light stream field picture for receiving；Convolutional layer is will to input
The received light stream field picture of layer forms the process of characteristic image by convolution kernel, and N number of input feature vector is converted M (M > by convolutional layer
N) a output characteristic pattern is connected between layer neuron, as weight parameter matrix, the generally square matrix of K*K, K using convolution kernel
For positive integer, and every characteristic image shares same weight parameter matrix and biasing.It is slided in the input picture of this layer
Convolution kernel by the weight weighted sum in convolution kernel region and convolution kernel and is added amount of bias and passes through ReLu activation primitive again
Form one unit of characteristic image of output.Convolution kernel progress convolution operation in same convolutional layer using identical size obtains more
A characteristic image, but the weight parameter different from of each convolution kernel, therefore can more fully extract feature.In order to make convolution
Core adapts to the size of image, and convolution kernel size used in different convolutional layers is reduced with the diminution of input feature vector image, from
And the precision of entire model is improved, and convolution nuclear volume used in each convolutional layer is all different under normal circumstances.Entire mistake
Journey is expressed by formula：
Wherein l indicates currently be which layer, and j is the quantity of current layer convolution kernel, that is, exports the quantity of image, and N is input
To the amount of images of this layer,Indicate the value of a certain unit in l j-th of characteristic image of layer,I-th is indicated in one layer
Corresponding matrix in image,Indicate j-th of convolution nuclear matrix of l layer, each layer has unique biasing BlAnd ReLu
Activation primitive f.
Further, activation primitive used in convolutional layer is ReLu function, and convergence rate is than in traditional neural network
Sigmoid activation primitive it is many fastly.Compared to other activation primitives, ReLu only needs a threshold value to can be obtained by activation letter
Number, so that activation is simpler, the training of model and the speed of service are greatly increased.
Connection pool layer is required after each convolutional layer and carries out down-sampling, is operated only complete in the same characteristic image
At not influenced between different characteristic image, not changing characteristic image quantity but change the size of each characteristic image.Convolutional layer obtains
To M characteristic image be all used as the input of pond layer to carry out secondary feature extraction.The effect of pond layer be to image not
Feature progress aggregate statistics with position keep the feature of image more obvious to obtain the lower image of dimension.Pond layer is normal
Downsapling method has mean value pondization and maximum pond.The present embodiment is unified adopt to characteristic image using maximum pond
Sample operation, i.e., value of the maximum value of all units as output layer corresponding position neuron in down-sampling window.Assuming that maximum pond
The sliding matrix size of change is m*m, and the characteristic image exported after down-sampling reduces m times respectively on two dimensions, then will be defeated
Result does nonlinear transformation by activation primitive out, and entire pond process can be used following formula to express：
Y=f (max (xi)+B)
Wherein xiIndicate the value of certain unit i in some down-sampling window, B indicates amount of bias, and f indicates the ReLu of pond layer
Activation primitive, y indicate down-sampling after image in corresponding unit value.
There are multilayer convolutional layer and pond layer in depth convolutional neural networks, remaining convolutional layer and pond layer and aforesaid operations one
It causes.Upper one layer of pond layer connects next layer of convolutional layer, and feature extraction is repeated, because of referred to herein as depth convolution.All convolution
After pondization operation, the last layer pond layer exports final characteristic image to full articulamentum, and the effect of full articulamentum is to connect
All features are connect, are responsible for feature being conveyed to output layer；Output layer is softmax classifier, is determined according to final classification number
Neuron number in classifier, finally obtains classification results, judges whether current video image occurs fire.
Step 4：Backpropagation optimizes the depth convolutional neural networks model of more size convolution kernels.Using training set to foundation
Good model is trained, using the optimization method of backpropagation successively to the weight parameter and bias arrived used in every layer
Reasonably optimizing is carried out, higher precision and better prediction effect can be reached.Feature of the error signal from sub-sampling layer
The characteristic pattern for scheming convolutional layer forward is propagated, and the optimization of parameter is completed by a convolution process.But the precision of prediction of training set
It is not easy excessively high, avoids the effect of over-fitting as far as possible, better effect can be played in practical applications.
So far, the foundation for completing the depth convolutional neural networks of entire more size convolution kernels, after training and has higher
The model of precision can be used in actual system host, and it is pre- that the image data that network high-definition camera is acquired back carries out fire
It surveys.
Present invention uses improved convolutional neural networks, have reached accurate extraction feature using various sizes of convolution kernel
Effect, the optical flow field by forming video image is more convenient for showing the position of moving object and form in image, then made
For the input of model, so predict in image with the presence or absence of flame, cigarette and other may cause the substance of fire, compared to
Other fire prediction methods, model prediction accuracy established by the present invention is higher, predicts faster.The present invention adds to a certain extent
Fast fire discovery speed, the use for reducing manpower and material resources, and it is reasonable to be supplied to administrative department according to fire development situation
Fire extinguishing strategy, economic loss and casualties is effectively reduced.
Claims (8)
1. a kind of forest fire detection method of the depth convolution model based on more size convolution kernels, it is characterised in that including as follows
Step：
【1】Acquire the video image of forest；
【2】Using optical flow method, generate a width light stream field picture per adjacent two field pictures, and by the optical flow field image set of generation press than
Example is divided into training set and test set；
【3】The depth convolutional neural networks model for establishing more size convolution kernels, by step【2】Obtained training set is as nerve net
The input of network model, is trained model, by step【2】Obtained test set is used for the precision of prediction of testing model, constantly
Optimized model；
【4】The light stream field picture for generating region to be predicted, detects Fires Occurred using the model after optimization.
2. the forest fire detection method of the depth convolution model as described in claim 1 based on more size convolution kernels, special
Sign is：Using the video image of network high-definition camera acquisition forest, control is transferred data to by wireless bridge structure
Center.
3. the forest fire detection method of the depth convolution model as described in claim 1 based on more size convolution kernels, special
Sign is：The light of each pixel on adjacent two frame video image is calculated using the Horn-Schunck algorithm based on gradient
Stream, obtains the light stream field picture of adjacent two field pictures.
4. the forest fire detection method of the depth convolution model as described in claim 1 based on more size convolution kernels, special
Sign is：The depth convolutional neural networks model of more size convolution kernels includes input layer, several convolutional layers, several pond layers, complete
Articulamentum and output layer；The input layer is for receiving light stream field picture；The convolutional layer is used for the received light of input layer
Flow field image carries out a feature extraction by convolution kernel；Each convolutional layer connects a pond layer and carries out down-sampling, convolutional layer
Obtained characteristic image is all used as the input of pond layer to carry out Further Feature Extraction, and pond layer is to characteristic image different location
Feature carries out aggregate statistics, to obtain the lower image of dimension；Upper one layer of pond layer is connect with next layer of convolutional layer, repeatedly into
Row feature extraction, to the last one layer of pond layer exports final characteristic image to full articulamentum；Full articulamentum connects all features,
And it is conveyed to output layer and classifies；Output layer finally obtains classification results using classifier.
5. the forest fire detection method of the depth convolution model as claimed in claim 4 based on more size convolution kernels, special
Sign is：Same convolutional layer carries out convolution operation using the convolution kernel of identical size and obtains multiple characteristic images, each convolutional layer
Used convolution kernel size changes with the size of input feature vector image, and each convolution kernel is weight parameter matrix, and every
Characteristic image shares same weight parameter matrix and biasing.
6. the forest fire detection method of the depth convolution model as claimed in claim 4 based on more size convolution kernels, special
Sign is：Pond layer uses maximum pond method.
7. the forest fire detection method of the depth convolution model as claimed in claim 4 based on more size convolution kernels, special
Sign is：All convolutional layers and pond layer are all made of ReLu activation primitive, provide model robustness, accelerate model convergence.
8. the forest fire detection method of the depth convolution model as described in claim 1 based on more size convolution kernels, special
Sign is：Using weight parameter and the value of biasing in BackForward backpropagation modification model, model accuracy is improved；Mistake
Signal is propagated from the characteristic pattern of the characteristic pattern of sub-sampling layer convolutional layer forward, completes the excellent of parameter by a convolution process
Change.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584066.8A CN108921039A (en) | 2018-06-07 | 2018-06-07 | The forest fire detection method of depth convolution model based on more size convolution kernels |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584066.8A CN108921039A (en) | 2018-06-07 | 2018-06-07 | The forest fire detection method of depth convolution model based on more size convolution kernels |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108921039A true CN108921039A (en) | 2018-11-30 |
Family
ID=64420280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810584066.8A Pending CN108921039A (en) | 2018-06-07 | 2018-06-07 | The forest fire detection method of depth convolution model based on more size convolution kernels |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921039A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785574A (en) * | 2019-01-21 | 2019-05-21 | 五邑大学 | A kind of fire detection method based on deep learning |
| CN109815904A (en) * | 2019-01-25 | 2019-05-28 | 吉林大学 | Fire identification method based on convolutional neural network |
| CN110051954A (en) * | 2019-04-19 | 2019-07-26 | 辽宁科技大学 | A kind of split type Super High full-automatic fire-extinguishing machine people and its control method |
| CN110147758A (en) * | 2019-05-17 | 2019-08-20 | 电子科技大学成都学院 | A kind of forest fire protection method based on deep learning |
| CN110222828A (en) * | 2019-06-12 | 2019-09-10 | 西安交通大学 | A kind of Unsteady Flow method for quick predicting based on interacting depth neural network |
| CN110379118A (en) * | 2019-07-26 | 2019-10-25 | 中车青岛四方车辆研究所有限公司 | Fire prevention intelligent monitor system and method under train vehicle |
| CN111242895A (en) * | 2019-12-31 | 2020-06-05 | 福建工程学院 | Bamboo chip wormhole and mildew spot detection method based on convolution flexible neural forest |
| CN111539325A (en) * | 2020-04-23 | 2020-08-14 | 四川旅游学院 | Forest fire detection method based on deep learning |
| WO2020181662A1 (en) * | 2019-03-11 | 2020-09-17 | 北京大学 | Monitoring method and system for protecting privacy |
| CN112309068A (en) * | 2020-10-29 | 2021-02-02 | 电子科技大学中山学院 | Forest fire early warning method based on deep learning |
| CN112488213A (en) * | 2020-12-03 | 2021-03-12 | 杭州电子科技大学 | Fire picture classification method based on multi-scale feature learning network |
| CN112881637A (en) * | 2021-02-08 | 2021-06-01 | 东华大学 | Visual osmotic membrane treatment wastewater detection device |
| CN113012383A (en) * | 2021-03-26 | 2021-06-22 | 深圳市安软科技股份有限公司 | Fire detection alarm method, related system, related equipment and storage medium |
| CN113101579A (en) * | 2021-05-12 | 2021-07-13 | 广东喷保消防科技有限公司 | Intelligent fire extinguishing equipment based on combination of water mist pump and AR |
| CN114648853A (en) * | 2022-03-09 | 2022-06-21 | 国网安徽省电力有限公司电力科学研究院 | Early fire mode identification and grading early warning system of high-voltage switch cabinet |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794486A (en) * | 2015-04-10 | 2015-07-22 | 电子科技大学 | Video smoke detecting method based on multi-feature fusion |
| 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 |
-
2018
- 2018-06-07 CN CN201810584066.8A patent/CN108921039A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794486A (en) * | 2015-04-10 | 2015-07-22 | 电子科技大学 | Video smoke detecting method based on multi-feature fusion |
| 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 |
Non-Patent Citations (4)
| Title |
|---|
| ENRIC MEINHARDT-LLOPIS 等: ""Horn-Schunck Optical Flow with a Multi-scale Strategy"", 《IMAGE PROCESSING ON LINE》 * |
| 于春雨: ""基于光流法火灾烟雾视频图像识别及多信息融合探测算法研究"", 《中国博士学位论文全文数据库信息科技辑》 * |
| 孙健: ""基于视频内容分析的火灾烟雾检测算法研究与实现"", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
| 赵小川: "《MATLAB图像处理——程序实现与模块化仿真》", 31 January 2014 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785574A (en) * | 2019-01-21 | 2019-05-21 | 五邑大学 | A kind of fire detection method based on deep learning |
| CN109815904A (en) * | 2019-01-25 | 2019-05-28 | 吉林大学 | Fire identification method based on convolutional neural network |
| CN109815904B (en) * | 2019-01-25 | 2022-05-13 | 吉林大学 | Fire identification method based on convolutional neural network |
| WO2020181662A1 (en) * | 2019-03-11 | 2020-09-17 | 北京大学 | Monitoring method and system for protecting privacy |
| CN110051954A (en) * | 2019-04-19 | 2019-07-26 | 辽宁科技大学 | A kind of split type Super High full-automatic fire-extinguishing machine people and its control method |
| CN110147758A (en) * | 2019-05-17 | 2019-08-20 | 电子科技大学成都学院 | A kind of forest fire protection method based on deep learning |
| CN110222828A (en) * | 2019-06-12 | 2019-09-10 | 西安交通大学 | A kind of Unsteady Flow method for quick predicting based on interacting depth neural network |
| CN110379118A (en) * | 2019-07-26 | 2019-10-25 | 中车青岛四方车辆研究所有限公司 | Fire prevention intelligent monitor system and method under train vehicle |
| CN111242895A (en) * | 2019-12-31 | 2020-06-05 | 福建工程学院 | Bamboo chip wormhole and mildew spot detection method based on convolution flexible neural forest |
| CN111242895B (en) * | 2019-12-31 | 2023-04-18 | 福建工程学院 | Bamboo chip wormhole and mildew detection method based on convolution flexible neural forest |
| CN111539325A (en) * | 2020-04-23 | 2020-08-14 | 四川旅游学院 | Forest fire detection method based on deep learning |
| CN112309068A (en) * | 2020-10-29 | 2021-02-02 | 电子科技大学中山学院 | Forest fire early warning method based on deep learning |
| CN112488213A (en) * | 2020-12-03 | 2021-03-12 | 杭州电子科技大学 | Fire picture classification method based on multi-scale feature learning network |
| CN112881637A (en) * | 2021-02-08 | 2021-06-01 | 东华大学 | Visual osmotic membrane treatment wastewater detection device |
| CN113012383A (en) * | 2021-03-26 | 2021-06-22 | 深圳市安软科技股份有限公司 | Fire detection alarm method, related system, related equipment and storage medium |
| CN113012383B (en) * | 2021-03-26 | 2022-12-30 | 深圳市安软科技股份有限公司 | Fire detection alarm method, related system, related equipment and storage medium |
| CN113101579A (en) * | 2021-05-12 | 2021-07-13 | 广东喷保消防科技有限公司 | Intelligent fire extinguishing equipment based on combination of water mist pump and AR |
| CN113101579B (en) * | 2021-05-12 | 2021-12-07 | 广东喷保消防科技有限公司 | Intelligent fire extinguishing equipment based on combination of water mist pump and AR |
| CN114648853A (en) * | 2022-03-09 | 2022-06-21 | 国网安徽省电力有限公司电力科学研究院 | Early fire mode identification and grading early warning system of high-voltage switch cabinet |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108921039A (en) | The forest fire detection method of depth convolution model based on more size convolution kernels | |
| WO2020173226A1 (en) | Spatial-temporal behavior detection method | |
| EP3614308B1 (en) | Joint deep learning for land cover and land use classification | |
| Tao et al. | Smoke detection based on deep convolutional neural networks | |
| Qin et al. | Applying the convolutional neural network deep learning technology to behavioural recognition in intelligent video | |
| CN104992223B (en) | Intensive Population size estimation method based on deep learning | |
| CN104166861B (en) | A kind of pedestrian detection method | |
| CN110378259A (en) | A kind of multiple target Activity recognition method and system towards monitor video | |
| CN110070530A (en) | A kind of powerline ice-covering detection method based on deep neural network | |
| CN107016357A (en) | A kind of video pedestrian detection method based on time-domain convolutional neural networks | |
| CN104320617B (en) | A kind of round-the-clock video frequency monitoring method based on deep learning | |
| CN107729799A (en) | Crowd's abnormal behaviour vision-based detection and analyzing and alarming system based on depth convolutional neural networks | |
| CN111047818A (en) | Forest fire early warning system based on video image | |
| CN106407903A (en) | Multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method | |
| CN108364097A (en) | Based on the typhoon cloud system prediction technique for generating confrontation network | |
| CN110378288A (en) | A kind of multistage spatiotemporal motion object detection method based on deep learning | |
| CN109508360A (en) | A kind of polynary flow data space-time autocorrelation analysis method of geography based on cellular automata | |
| CN109376637A (en) | Passenger number statistical system based on video monitoring image processing | |
| CN100565557C (en) | System for tracking infrared human body target based on corpuscle dynamic sampling model | |
| CN110135476A (en) | A kind of detection method of personal safety equipment, device, equipment and system | |
| CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
| CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
| CN110263654A (en) | A kind of flame detecting method, device and embedded device | |
| CN109935080A (en) | The monitoring system and method that a kind of vehicle flowrate on traffic route calculates in real time | |
| CN111046728A (en) | Straw combustion detection method based on characteristic pyramid 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181130 |