CN109887220A - Air conditioner and control method thereof - Google Patents
Air conditioner and control method thereof Download PDFInfo
- Publication number
- CN109887220A CN109887220A CN201910063794.9A CN201910063794A CN109887220A CN 109887220 A CN109887220 A CN 109887220A CN 201910063794 A CN201910063794 A CN 201910063794A CN 109887220 A CN109887220 A CN 109887220A
- Authority
- CN
- China
- Prior art keywords
- air
- fire source
- layer
- conditioning
- target environment
- 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
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000004378 air conditioning Methods 0.000 claims description 97
- 238000003062 neural network model Methods 0.000 claims description 46
- 230000004913 activation Effects 0.000 claims description 26
- 238000013527 convolutional neural network Methods 0.000 claims description 26
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000005070 sampling Methods 0.000 claims description 12
- 108010001267 Protein Subunits Proteins 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims description 2
- 230000000007 visual effect Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 13
- 238000012545 processing Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 210000002569 neuron Anatomy 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 239000000203 mixture Substances 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 230000002779 inactivation Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 230000002618 waking effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 229940040145 liniment Drugs 0.000 description 1
- 239000000865 liniment Substances 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The application provides an air conditioner and a control method of the air conditioner, wherein the air conditioner comprises: the image acquisition unit is used for acquiring a target environment image of an area where the air conditioner is located; the image recognition unit is used for determining whether a fire source is displayed in the target environment image; and the alarm unit is used for sending out reminding information when a fire source is displayed in the target environment image. The air conditioner in this application has the fire alarm function, consequently need not additionally to install fire alarm, has saved the space, simultaneously because need not install fire alarm on ceiling etc. position, so can not destroy indoor holistic aesthetic property to improve user's visual experience.
Description
Technical field
This application involves field of air conditioning, the in particular to control method of air-conditioning and air-conditioning.
Background technique
With the present improvement of people's living standards, air-conditioning has become the essential a part of people's daily life,
And with the improvement of people's safety consciousness, more and more families and public place are assembled with fire-alarm, existing skill
In art, need to install air-conditioning and fire-alarm respectively, two equipment occupy biggish space, and are arranged separately in ceiling
Fire-alarm often destroy aesthetics whole in house.
Summary of the invention
This application provides a kind of empty and air-conditioning control methods, for saving air-conditioning and the occupied sky of fire-alarm
Between, and improve the aesthetics of indoor entirety.
To solve the above-mentioned problems, as the one aspect of the application, a kind of air-conditioning is provided, comprising:
Image acquisition units, for acquiring the target environment image of air-conditioning region;
Image identification unit, for determining in target environment image whether show fire source;
Alarm unit issues prompting message when for showing fire source in target environment image.
Optionally, image identification unit is used to determine whether show in the target environment image using neural network model
There is fire source, wherein the target environment image is the input value of the neural network model.
Optionally, neural network model is appointing in depth convolutional neural networks, residual error neural network or BP neural network
Meaning one.
Optionally, neural network model is depth convolutional neural networks;
Depth convolutional neural networks include sequentially connected input layer, N number of convolutional encoding network, N number of deconvolution decoding net
Network and output layer;
Convolutional encoding network includes the first convolutional layer, first regularization layer, the first activation primitive layer and maximum pond layer;
Deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and the second activation primitive layer;
Wherein, N is more than or equal to 1.
Optionally, N is equal to 16;
And/or convolutional encoding network further include: the inactivation between the first activation primitive layer and maximum pond layer is set
Layer.
Optionally, image identification unit includes: classification subelement and judgment sub-unit;
Classification subelement, for determining the pixel class of each pixel in target environment image, wherein pixel class includes
Fire source pixel and non-fire source pixel.
Judgment sub-unit, for determining mesh when the fire source pixel ratio in target environment image is greater than the first preset ratio
Fire source is shown in mark ambient image;
Wherein, fire source pixel ratio is the ratio of the number of pixels of fire source pixel and total number of pixels of target environment image
Value.
Optionally, alarm unit is specifically used for:
When fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to
The terminal of air-conditioning binding sends prompting message;And/or
Prompting message is issued when fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to
The location information of prompting message and air-conditioning is sent with the server of air-conditioning binding, so that server control is pre- apart from being less than with air-conditioning
If the smart machine of distance issues prompting message;And/or
When fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to
The location information of fire department alert and air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
The application also proposes a kind of control method of air-conditioning, and air-conditioning includes image acquisition units, image identification unit and report
Alert unit characterized by comprising
The target environment image of image acquisition units acquisition air-conditioning region;
Image identification unit determines in target environment image whether show fire source;
Alarm unit issues prompting message when showing fire source in target environment image.
Optionally, image identification unit determines in target environment image whether show fire source using neural network model,
Wherein target environment image is the input value of neural network model.
Optionally, neural network model is appointing in depth convolutional neural networks, residual error neural network or BP neural network
Meaning one.
Optionally, neural network model is depth convolutional neural networks;
Depth convolutional neural networks include sequentially connected input layer, N number of convolutional encoding network, N number of deconvolution decoding net
Network and output layer;
Convolutional encoding network includes the first convolutional layer, first regularization layer, the first activation primitive layer and maximum pond layer;
Deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and the second activation primitive layer;
Wherein, N is more than or equal to 1.
Optionally, N is equal to 16;
And/or convolutional encoding network further include: the inactivation between the first activation primitive layer and maximum pond layer is set
Layer.
Optionally, image identification unit determines that fire source whether is shown in target environment image includes:
Determine the pixel class of each pixel in target environment image, wherein pixel class includes fire source pixel and non-fire
Source pixel.
It determines in target environment image and shows when the fire source pixel ratio in target environment image is greater than the first preset ratio
It is shown with fire source;
Wherein, fire source pixel ratio is total pixel of number of pixels and target environment image that pixel class is fire source pixel
The ratio of number.
Optionally, prompting message is issued when alarm unit has fire source in target environment image, comprising:
When fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to
The terminal of air-conditioning binding sends prompting message;And/or
Prompting message is issued when fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to
The location information of prompting message and air-conditioning is sent with the server of air-conditioning binding, so that server control is pre- apart from being less than with air-conditioning
If the smart machine of distance issues prompting message;And/or
When fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to
The location information of fire department alert and air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
Present applicant proposes the control methods of a kind of air-conditioning and air-conditioning, and wherein air-conditioning has fire alarm function, pass through bat
It takes the photograph target environment image and identification is carried out to target environment image and judge whether there is fire source, prompting message is issued when there is fire source,
There is no need to additionally install fire-alarm, space is saved, simultaneously as not having to install fire report on the positions such as ceiling
Alert device, so the aesthetics of indoor entirety will not be destroyed, so as to improve the visual experience of user.
Detailed description of the invention
Fig. 1 is a kind of composition figure of air-conditioning in the embodiment of the present application;
Fig. 2 is a kind of composition figure of depth convolutional neural networks in the embodiment of the present application;
Fig. 3 is a kind of composition figure of convolutional encoding network in the embodiment of the present application;
Fig. 4 is a kind of composition figure of deconvolution decoding network in the embodiment of the present application;
Fig. 5 is a kind of composition figure of image identification unit in the embodiment of the present application;
Fig. 6 is a kind of flow chart of the control method of air-conditioning in the embodiment of the present application.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with the application specific embodiment and
Technical scheme is clearly and completely described in corresponding attached drawing.Obviously, described embodiment is only the application one
Section Example, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall in the protection scope of this application.
It should be noted that the description and claims of this application and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to embodiments herein described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that cover
Covering non-exclusive includes to be not necessarily limited to for example, containing the process, method of a series of steps or units, device, product or air-conditioning
Step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, product
Or other step or units that air-conditioning is intrinsic.
In the prior art, air-conditioning and fire-alarm are two independent devices, install air-conditioning and fire alarm respectively
Device needs to occupy biggish space, and fire-alarm is often ceiling mounted the aesthetics for destroying indoor entirety,
To solve the above-mentioned problems, as shown in Figure 1, present applicant proposes a kind of air-conditionings, comprising:
Image acquisition units 10, for acquiring the target environment image of air-conditioning region;
Image identification unit 20, for determining in target environment image whether show fire source;
Alarm unit 30 issues prompting message when for showing fire source in target environment image.
Specifically, image acquisition units 10 for example can be the camera being arranged on air-conditioning, it is preferably provided with multiple camera shootings
Head, thus the target environment image of omnidirectional shooting air-conditioning region, air-conditioning region refers to the position area of air-conditioning installation
Domain, such as air-conditioning are mounted on parlor, then air-conditioning region is exactly parlor.Image identification unit 20 is true using image processing techniques
Whether fire source is shown in the ambient image that sets the goal, when showing fire source, alarm unit 30 issues prompting message, such as alarms
Unit 30 may include buzzer, and buzzer issues prompting message by way of sounding an alarm sound.In the present embodiment, air-conditioning
Space is saved there is no need to additionally install fire-alarm with fire alarm function, simultaneously as not having on the ceiling
Fire-alarm is installed, so the aesthetics of indoor entirety will not be destroyed, so as to improve the visual experience of user.
In some embodiments, described image recognition unit is used to determine the target environment figure using neural network model
Whether fire source is shown as in, wherein the target environment image is the input value of the neural network model.
Specifically, target environment image is the input value of neural network model, it is stored in image identification unit 10 in advance
The neural network model of foundation pre-enters a large amount of image and first provides recognition result when establishing neural network model, knows
Other result includes that fire source whether is shown in image, is then trained to neural network model, to establish image and identification
As a result the connection between determines each middle layer of neural network model, in the present embodiment, the output of neural network model
Value includes whether showing fire source in the target environment image.Optionally, neural network model be depth convolutional neural networks,
Any one in residual error neural network or BP neural network.
In some alternative embodiments, neural network model is depth convolutional neural networks;
Depth convolutional neural networks as shown in Figure 2 include sequentially connected input layer, N number of convolutional encoding network, N number of warp
Product decoding network and output layer;
As shown in figure 3, convolutional encoding network include the first convolutional layer, first regularization layer, the first activation primitive layer and
Maximum pond layer;
As shown in figure 4, deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and second
Activation primitive layer;Wherein, N is more than or equal to 1.
Specifically, in the present embodiment, input layer is for inputting target environment image, convolutional encoding network and deconvolution solution
Code network is for handling target environment image, and output layer is for input processing as a result, processing result includes target environment
Whether include fire source in image, in the present embodiment in depth convolutional neural networks model using N layer convolutional encoding network with
It is corresponding with the deconvolution coding network of the identical number of plies, wherein N is preferably equal to 16.When N is preferably not less than 2, depth convolutional Neural
Network includes multilayer convolutional encoding network and multilayer deconvolution decoding network, it should be noted that multilayer convolution is compiled in this implementation
Not using the full articulamentum in traditional neural network in code network, be conducive to the convolutional encoding in bottommost layer (n-th layer) in this way
High-resolution characteristic pattern is exported in network, and reduces the parameter of convolutional encoding network, to reduce training convolutional coding
The training time of network.When N is not less than 2, the input of first layer convolutional encoding network is target environment image, is located at first layer
The input value of the convolutional encoding network of other layers after convolutional encoding network in an encoding process is all upper one layer of convolutional encoding
The output characteristic pattern of network, the first convolutional layer are to carry out feature extraction (first layer to output characteristic pattern by 3 × 3 convolution kernel
Convolutional encoding network is to carry out feature extraction to target environment image), then first regularization layer carries out the feature of extraction
Regularization operation (Batch Normalization) is criticized, then the first activation primitive layer (preferably uses ReLu using activation primitive
Activation primitive) Nonlinear Mapping is carried out to the feature after batch regularization operation, finally carry out maximum pondization processing output characteristic pattern
Picture, maximum pondization handles translation invariance of the available target environment image in the variation of small space displacement, because N is not less than
2, that is, multiple maximum pondization processing has been carried out, and repeatedly maximum pondization processing can obtain the characteristic pattern with higher robustness
Picture.But it is continuous to carry out pond down-sampling, it will cause target environment image and be constantly distorted, boundary information is lost, and in the application
It is preferred that needing to carry out the pixel of target environment image pixel class classification to judge whether display or fire source, in pixel class
Not Fen Lei when need to be split pixel, and boundary information is lost and is unfavorable for segmentation to target environment image.So
It is corresponding, it is provided with deconvolution decoding network, after convolutional encoding network in this application to restore in target environment image
Information.Optionally, in order to restore the information in target environment image as far as possible, maximum pond Hua Chu is carried out in maximum pond layer
The aspect indexing of maximum eigenvalue is recorded during reason.During deconvolution coding network is decoded, up-sampling layer is used
The aspect indexing recorded in maximum pond treatment process up-samples input feature vector, and the second convolutional layer uses a convolution kernel
Convolution operation is carried out to the sparse features figure that up-sampling obtains and obtains dense characteristic pattern, then second batch regularization layer is to dense
Characteristic pattern carry out batch Regularization, then the second activation primitive layer carries out non-linear reflects to the feature after batch Regularization
It penetrates and exports characteristic image.In the present embodiment, batch regularization has all been correspondingly arranged after the first convolutional layer and the second convolutional layer
Layer, is difficult to trained disadvantage and accelerates the training process of neural network to overcome depth convolutional neural networks, it is therefore prevented that
The gradient disappearance problem that depth convolutional neural networks are easy to appear in the training process, help to improve trained convergence rate and
Model accuracy.
Optionally, convolutional encoding network further include be arranged between the first activation primitive layer and maximum pond layer can inactivation
Layer (the DropOut layer i.e. in field of neural networks), the purposes of deactivating layer is to prevent over-fitting, improves neural network model
Generalization ability, when designing neural network model, it is (i.e. several that every layer of neuron of setting represents the intermediate features that one learns
The combination of weight), all neuron collective effect characterizes input data (target environment image) in neural network model
Particular community.During training neural network model, when input data is too small for the complexity relative to network,
It just will appear over-fitting, it is clear that the feature that at this moment each neuron indicates has many repetitions and redundancy between each other.Deactivating layer
Directly effect is the quantity of reduction intermediate features increases the orthogonality between every layer of each feature to reduce redundancy.
In some embodiments, as shown in figure 5, image identification unit 20 includes: classification subelement 21 and judgment sub-unit
22;
Classification subelement 21, for determining the pixel class of each pixel in target environment image, wherein pixel class packet
Include fire source pixel and non-fire source pixel.Judgment sub-unit 22 is greater than for the fire source pixel ratio in the target environment image
It is determined when one preset ratio in target environment image and shows fire source;
Wherein, fire source pixel ratio is the ratio of the number of pixels of fire source pixel and total number of pixels of target environment image
Value.
Specifically, any picture is all made of limited pixel, wherein when showing fire source in target environment image,
The pixel for showing fire source is exactly fire source pixel, other pixels are just non-fire source pixels, when image identification unit uses neural network
When model determines whether show fire source in target environment image, neural network model be can store in classification subelement 21,
It needs to be trained neural network model after establishing neural network model, when training, first acquires some pictures, picture point
For comprising fire source and not fire source, and manual mark is carried out to the picture of liniment, each pixel is divided into 2 kinds of pixel classes
Other: fire source pixel, non-fire source pixel read data in order to facilitate computer, are respectively labeled as 0 (non-fire source pixel) and 1 (fire source
Pixel), then start to train, specific training method can use any existing neural network training method, the application couple
This is not construed as limiting.The output layer of the neural network trained exports the pixel class and fire source pixel ratio of each pixel, from
And determine in target environment image whether show fire source.
It is shot it should be noted that the number of image acquisition units 10 can be multiple, different image acquisition units 10
Region it is different, so the fire source shown in the target environment image that different image acquisition units are shot and practical fire source is big
Small scale is also different, and therefore, the first ratio is mutually bound with image acquisition units, different image acquisition units corresponding
One ratio is different.The purpose that the first ratio is arranged is to prevent from judging by accident, because user is cooked or used lighter using gas-cooker
Also fire source can be generated when lighting a cigarette, but these are not the fire source of meeting fire, and alarm should not be issued when detecting these fire sources
Information.
Optionally, alarm unit 30 is specifically used for:
When fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to
The terminal of air-conditioning binding sends prompting message;And/or
Prompting message is issued when fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to
The location information of prompting message and air-conditioning is sent with the server of air-conditioning binding, so that server control is pre- apart from being less than with air-conditioning
If the smart machine of distance issues prompting message;And/or
When fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to
The location information of fire department alert and air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
Alarm unit 30 does not make movement if without fire source, when showing fire source, the bigger table of fire source pixel ratio
Open fire gesture is bigger, then judges to intensity of a fire size if there is fire source, and determine to mention using which kind of according to intensity of a fire size
The mode of waking up.It should be noted that the first ratio, the second ratio, third ratio and the 4th ratio are all and image acquisition units phase
Binding, different image acquisition units are because the region of shooting is different, corresponding first ratio, the second ratio, third
Ratio and the 4th ratio difference.
Optional alarm unit is specifically used for, and when detecting the target environment image for showing fire source for the first time, uses
The sound prompting of buzzer and to user mobile phone send alarm.When continuous collecting is to the target environment image for showing fire source, and
When continuing the sound prompting of first time nobody releasing buzzer, air-conditioning can send prompting message and air-conditioning to remote server
Location information, server receives and can send control to the smart machine on air-conditioning periphery after prompting message and location information and refer to
It enables, control smart machine sounds an alarm.When continuous collecting is rendered as increasing to the target environment image and the intensity of a fire for showing fire source
Big trend, and continue second time nobody release buzzer sound prompting when, air-conditioning automatically dials Fire telephone, and
Inform fire scene.
As shown in fig. 6, the application also proposes a kind of control method of air-conditioning, air-conditioning includes image acquisition units, image knowledge
Other unit and alarm unit characterized by comprising
The target environment image of image acquisition units acquisition air-conditioning region;
Image identification unit determines in target environment image whether show fire source;
Alarm unit issues prompting message when showing fire source in target environment image.
Air-conditioning in the control method of air-conditioning during the application is any can be any one air-conditioning of the application proposition,
That is the control method of the air-conditioning of the application proposition can be any air-conditioning proposed for the application.The control that the application proposes
Method is by acquisition target environment image and identifies in target environment image whether there is fire source, so that air-conditioning can be known
Other fire source, and can automatically be sounded an alarm according to the situation of fire source, so that air-conditioning is more intelligent, improves user and use air-conditioning
When safety coefficient.
Optionally, described image recognition unit determines whether show in the target environment image using neural network model
There is fire source, wherein the target environment image is the input value of the neural network model.Specifically, in image identification unit 10
It is stored with the neural network model pre-established, when establishing neural network model, a large amount of image is pre-entered and first provides
Recognition result, recognition result include that fire source whether is shown in image, are then trained to neural network model, to establish
Contacting between image and recognition result, the method for training neural network model can be any existing method, the application couple
This is not construed as limiting, and in the present embodiment, whether the output valve of neural network model includes showing in the target environment image
Fire source.Optionally, neural network model is any one in depth convolutional neural networks, residual error neural network or BP neural network
It is a.Optionally, neural network model is any one in depth convolutional neural networks, residual error neural network or BP neural network
It is a.
Optionally, neural network model is depth convolutional neural networks;
Depth convolutional neural networks include sequentially connected input layer, N number of convolutional encoding network, N number of deconvolution decoding net
Network and output layer;Any convolutional encoding network includes the first convolutional layer, first regularization layer, the first activation primitive layer and maximum
Pond layer;Deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and the second activation primitive layer;
Wherein, N is more than or equal to 1.N can also be more than or equal to 2, and preferably N is equal to 16.
Specifically, in the present embodiment, using N layers of convolutional encoding network and right in depth convolutional neural networks model
There should be the deconvolution coding network of the identical number of plies, when N is not less than 2, depth convolutional neural networks include multilayer convolutional encoding net
Network and multilayer deconvolution decoding network, it should be noted that without using tradition mind in multilayer convolutional encoding network in this implementation
Through the full articulamentum in network, be conducive to export high-resolution characteristic pattern in the convolutional encoding network of bottommost layer in this way, and
And reduce the parameter of convolutional encoding network, to reduce the training time of convolutional encoding network.When N is not less than 2, first layer
The input of convolutional encoding network is target environment image, the convolutional encoding of other layers after first layer convolutional encoding network
In an encoding process, the first convolutional layer is all the convolution kernel by 3 × 3 to the defeated of upper one layer of convolutional encoding network to network each time
Feature extraction is carried out out, and then first regularization layer carries out crowd regularization operation (Batch to the feature of extraction
Normalization), then the first activation primitive layer utilizes activation primitive (preferably using ReLu activation primitive) to batch regularization
Feature after operation carries out Nonlinear Mapping, finally carries out maximum pondization processing, and maximum pondization handles available target environment
Translation invariance of the image in the variation of small space displacement has carried out multiple maximum pondization processing because N is not less than 2, and more
Secondary maximum pondization processing can obtain the feature with higher robustness.But it is continuous to carry out pond down-sampling, it will cause mesh
Mark ambient image is constantly distorted, and boundary information is lost, and is unfavorable for the segmentation task to target environment image.So corresponding
, it is provided with deconvolution decoding network, after convolutional encoding network in this application to restore the information in target environment image.
Optionally, in order to restore the information in target environment image as far as possible, the mistake of maximum pondization processing is carried out in maximum pond layer
The aspect indexing of maximum eigenvalue is recorded in journey.During deconvolution coding network is decoded, up-sampling layer uses pond Hua Chu
The aspect indexing recorded during reason up-samples input feature vector to obtain sparse features figure, and the second convolutional layer uses a volume
The sparse features figure that product verification up-sampling obtains carries out convolution operation and obtains dense characteristic pattern, and then second batch regularization layer is right
Dense characteristic pattern carries out batch Regularization, and then the second activation primitive layer carries out the feature after batch Regularization non-thread
Property mapping.In the present embodiment, batch regularization layer has all been correspondingly arranged after the first convolutional layer and the second convolutional layer, to overcome
Depth convolutional neural networks are difficult to trained disadvantage and accelerate the training process of neural network, it is therefore prevented that depth convolutional Neural
The gradient disappearance problem that network is easy to appear in the training process, helps to improve trained convergence rate and model accuracy.
Optionally, the convolutional encoding network further include: setting the first activation primitive layer and maximum pond layer it
Between deactivating layer (the DropOut layer i.e. in field of neural networks), the purposes of deactivating layer is to prevent over-fitting, improves nerve net
The generalization ability of network model, when designing neural network model, every layer of neuron of setting represents the intermediate spy learnt
It levies (combinations of i.e. several weights), all neuron collective effect characterizes input data (target ring in neural network model
Border image) particular community.During training neural network model, number is inputted for the complexity relative to network
According to it is too small when, just will appear over-fitting, it is clear that the feature that at this moment each neuron indicates has many repetitions and redundancy between each other.
The direct effect of deactivating layer is the quantity of reduction intermediate features increases between every layer of each feature just to reduce redundancy
The property handed over.
Optionally, image identification unit determines that fire source whether is shown in target environment image includes:
Determine the pixel class of each pixel in target environment image, wherein pixel class includes fire source pixel and non-fire
Source pixel.
It determines in target environment image and shows when the fire source pixel ratio in target environment image is greater than the first preset ratio
It is shown with fire source;
Wherein, fire source pixel ratio is total pixel of number of pixels and target environment image that pixel class is fire source pixel
The ratio of number.
Optionally, prompting message is issued when alarm unit has fire source in target environment image, comprising:
When fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to
The terminal of air-conditioning binding sends prompting message;And/or
Prompting message is issued when fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to
The location information of prompting message and air-conditioning is sent with the server of air-conditioning binding, so that server control is pre- apart from being less than with air-conditioning
If the smart machine of distance issues prompting message;And/or
When fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to
The location information of fire department alert and air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
Optional alarm unit issues prompting message and specifically includes: detecting the target environment for showing fire source for the first time
When image, alarm is sent using the sound prompting of buzzer and to user mobile phone.When continuous collecting to the target for showing fire source
Ambient image, and continue first time nobody release buzzer sound prompting when, air-conditioning can to remote server send mention
The location information for the information and air-conditioning of waking up, server can be to the smart machines on air-conditioning periphery after receiving prompting message and location information
Control instruction is sent, control smart machine sounds an alarm.When continuous collecting to the target environment image and fire for showing fire source
Gesture be rendered as increase trend, and continue second time nobody release buzzer sound prompting when, air-conditioning is automatically dialed
Fire telephone, and inform fire scene.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (14)
1. a kind of air-conditioning characterized by comprising
Image acquisition units, for acquiring the target environment image of the air-conditioning region;
Image identification unit, for determining in the target environment image whether show fire source;
Alarm unit issues prompting message when for showing fire source in the target environment image.
2. air-conditioning according to claim 1, which is characterized in that described image recognition unit is used to use neural network model
It determines in the target environment image and whether shows fire source, wherein the target environment image is the neural network model
Input value.
3. air-conditioning according to claim 2, which is characterized in that
The neural network model is any one in depth convolutional neural networks, residual error neural network or BP neural network.
4. air-conditioning according to claim 3, which is characterized in that the neural network model is depth convolutional neural networks;
The depth convolutional neural networks include sequentially connected input layer, N number of convolutional encoding network, N number of deconvolution decoding net
Network and output layer;
The convolutional encoding network includes the first convolutional layer, first regularization layer, the first activation primitive layer and maximum pond layer;
The deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and the second activation primitive layer;
Wherein, N is more than or equal to 1.
5. air-conditioning according to claim 4, which is characterized in that
N is equal to 16;
And/or the convolutional encoding network further include: the mistake between the first activation primitive layer and maximum pond layer is set
Layer living.
6. air-conditioning according to claim 1-5, which is characterized in that described image recognition unit includes: class small pin for the case
Unit and judgment sub-unit;
The classification subelement, for determining the pixel class of each pixel in the target environment image, wherein the pixel
Classification includes fire source pixel and non-fire source pixel.
The judgment sub-unit, for true when the fire source pixel ratio in the target environment image is greater than the first preset ratio
Fire source is shown in the fixed target environment image;
Wherein, the fire source pixel ratio is the number of pixels of the fire source pixel and total pixel of the target environment image
Several ratio.
7. air-conditioning according to claim 6, which is characterized in that the alarm unit is specifically used for:
When the fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to
The terminal of the air-conditioning binding sends prompting message;And/or
Prompting message is issued when the fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to
The location information of prompting message and the air-conditioning is sent with the server of air-conditioning binding, so that server control and institute
State the smart machine sending prompting message that air-conditioning distance is less than pre-determined distance;And/or
When the fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to
The location information of fire department alert and the air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
8. a kind of control method of air-conditioning, the air-conditioning includes image acquisition units, image identification unit and alarm unit, spy
Sign is, comprising:
The target environment image of image acquisition units acquisition air-conditioning region;
Image identification unit determines in the target environment image whether show fire source;
Alarm unit issues prompting message when showing fire source in target environment image.
9. the control method of air-conditioning according to claim 8, which is characterized in that
Described image recognition unit determines in the target environment image whether show fire source using neural network model, wherein
The target environment image is the input value of the neural network model.
10. the control method of air-conditioning according to claim 9, which is characterized in that
Neural network model is any one in depth convolutional neural networks, residual error neural network or BP neural network.
11. the control method of air-conditioning according to claim 10, which is characterized in that neural network model is depth convolutional Neural net
Network;
Depth convolutional neural networks include sequentially connected input layer, N number of convolutional encoding network, N number of deconvolution decoding network and
Output layer;
The convolutional encoding network includes the first convolutional layer, first regularization layer, the first activation primitive layer and maximum pond layer;
The deconvolution decoding network includes up-sampling layer, the second convolutional layer, second batch regularization layer and the second activation primitive layer;
Wherein, N is more than or equal to 1.
12. the control method of air-conditioning according to claim 11, which is characterized in that
N is equal to 16;
And/or the convolutional encoding network further include: the mistake between the first activation primitive layer and maximum pond layer is set
Layer living.
13. according to the control method of the air-conditioning of any one of claim 8-13, which is characterized in that image identification unit determines target
Fire source whether is shown in ambient image includes:
Determine the pixel class of each pixel in target environment image, wherein pixel class includes fire source pixel and non-fire source picture
Element.
It determines in target environment image and shows when the fire source pixel ratio in target environment image is greater than the first preset ratio
Fire source;
Wherein, fire source pixel ratio is total number of pixels of number of pixels and target environment image that pixel class is fire source pixel
Ratio.
14. the control method of air-conditioning according to claim 13, which is characterized in that alarm unit has fire in target environment image
Prompting message is issued when source, comprising:
When fire source pixel ratio be greater than the first preset ratio and issue when less than the second preset ratio prompting message and to air-conditioning
The terminal of binding sends prompting message;And/or
Issue prompting message when fire source pixel ratio is greater than the second preset ratio and is less than third preset ratio, and to sky
Adjust the server of binding to send the location information of prompting message and air-conditioning so that server control and air-conditioning distance be less than it is default away from
From smart machine issue prompting message;And/or
When fire source pixel ratio is greater than third preset ratio and less than four preset ratios when issues prompting message, and to fire-fighting
The location information of department's alert and air-conditioning;
Wherein, the first preset ratio < second preset ratio < third preset ratio < the 4th preset ratio.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910063794.9A CN109887220A (en) | 2019-01-23 | 2019-01-23 | Air conditioner and control method thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910063794.9A CN109887220A (en) | 2019-01-23 | 2019-01-23 | Air conditioner and control method thereof |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109887220A true CN109887220A (en) | 2019-06-14 |
Family
ID=66926581
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910063794.9A Pending CN109887220A (en) | 2019-01-23 | 2019-01-23 | Air conditioner and control method thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109887220A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110896871A (en) * | 2019-09-30 | 2020-03-24 | 珠海格力电器股份有限公司 | Method and device for putting food and intelligent food throwing machine |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2000998A2 (en) * | 2007-05-31 | 2008-12-10 | Industrial Technology Research Institute | Flame detecting method and device |
CN103886344A (en) * | 2014-04-14 | 2014-06-25 | 西安科技大学 | Image type fire flame identification method |
CN203949288U (en) * | 2014-06-24 | 2014-11-19 | 广东美的暖通设备有限公司 | Air-conditioning system |
CN104200609A (en) * | 2014-09-27 | 2014-12-10 | 江阴市电工合金有限公司 | Electrotechnical alloy intelligent workshop |
CN104501363A (en) * | 2014-11-25 | 2015-04-08 | 珠海格力电器股份有限公司 | Air conditioner and control system thereof |
CN105336085A (en) * | 2015-09-02 | 2016-02-17 | 华南师范大学 | Remote large-space fire monitoring alarm method based on image processing technology |
CN106327777A (en) * | 2016-08-26 | 2017-01-11 | 珠海格力电器股份有限公司 | Air conditioner |
-
2019
- 2019-01-23 CN CN201910063794.9A patent/CN109887220A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2000998A2 (en) * | 2007-05-31 | 2008-12-10 | Industrial Technology Research Institute | Flame detecting method and device |
CN103886344A (en) * | 2014-04-14 | 2014-06-25 | 西安科技大学 | Image type fire flame identification method |
CN203949288U (en) * | 2014-06-24 | 2014-11-19 | 广东美的暖通设备有限公司 | Air-conditioning system |
CN104200609A (en) * | 2014-09-27 | 2014-12-10 | 江阴市电工合金有限公司 | Electrotechnical alloy intelligent workshop |
CN104501363A (en) * | 2014-11-25 | 2015-04-08 | 珠海格力电器股份有限公司 | Air conditioner and control system thereof |
CN105336085A (en) * | 2015-09-02 | 2016-02-17 | 华南师范大学 | Remote large-space fire monitoring alarm method based on image processing technology |
CN106327777A (en) * | 2016-08-26 | 2017-01-11 | 珠海格力电器股份有限公司 | Air conditioner |
Non-Patent Citations (1)
Title |
---|
吴宗胜等: "基于深度卷积神经网络的道路场景理解", 《计算机工程与应用》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110896871A (en) * | 2019-09-30 | 2020-03-24 | 珠海格力电器股份有限公司 | Method and device for putting food and intelligent food throwing machine |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110674790B (en) | Abnormal scene processing method and system in video monitoring | |
CN100464348C (en) | Video frequency monitoring, identification intelligont device and technical method | |
CN110827505A (en) | Smoke segmentation method based on deep learning | |
US11150056B2 (en) | System and method for dynamic camouflaging | |
CN105719439A (en) | Eye protection system and method | |
CN108334298A (en) | A kind of electronic label intelligent display system and method | |
CN108319892A (en) | A kind of vehicle safety method for early warning and system based on genetic algorithm | |
CN106354460A (en) | Physical interaction system and interaction method thereof | |
CN109887220A (en) | Air conditioner and control method thereof | |
CN107833193A (en) | A kind of simple lens global image restored method based on refinement network deep learning models | |
CN110740298A (en) | Distributed classroom discipline behavior detection system, method and medium | |
CN110956768A (en) | Automatic anti-theft device of intelligence house | |
CN115761618A (en) | Key site security monitoring image identification method | |
CN113963252A (en) | Safety helmet wearing early warning method and system based on image recognition | |
CN104184941A (en) | Monitoring system | |
CN106846522A (en) | A kind of calling device and the roll-call method for the device | |
CN109299660A (en) | A kind of Intelligent campus early warning platform | |
US20200380635A1 (en) | Configuring alpha channel video based on display settings | |
CN107329573A (en) | Control method and electronic equipment | |
CN112218414A (en) | Method and system for adjusting brightness of self-adaptive equipment | |
CN109099893B (en) | Monitoring method, medium and computing device used in movie theatre | |
CN106774010B (en) | A kind of movable-type intelligent steward system | |
CN107993388A (en) | A kind of Intelligent security apparatus, method and system | |
CN203966368U (en) | Building fire-fighting fire disaster escaping intelligent body check system | |
WO2017108466A1 (en) | Lighting systems and methods for crowd control |
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: 20190614 |