CN108764264A - Smog detection method, smoke detection system and computer installation - Google Patents
Smog detection method, smoke detection system and computer installation Download PDFInfo
- Publication number
- CN108764264A CN108764264A CN201810217094.6A CN201810217094A CN108764264A CN 108764264 A CN108764264 A CN 108764264A CN 201810217094 A CN201810217094 A CN 201810217094A CN 108764264 A CN108764264 A CN 108764264A
- Authority
- CN
- China
- Prior art keywords
- smog
- cnn models
- chimney
- training
- image data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The present invention proposes a kind of smog detection method, smoke detection system, computer installation and computer readable storage medium, wherein smog detection method:The image data collection of smog, chimney is obtained, and to image data collection point of addition label and class label;Image data collection with location tags and class label is input to as training sample set in CNN models and is trained, the CNN models after multiple training, and counting loss function while being trained are obtained;Smoke Detection is carried out to real-time video according to optimal CNN models are chosen in CNN models of the loss function after multiple training, to obtain Smoke Detection result.Smog detection method provided by the invention can timely detect smog in real-time video, can not only promote the accuracy rate of Smoke Detection, additionally it is possible to promote the detection speed of smog by identifying smog and chimney simultaneously in the video of big data quantity.
Description
Technical field
The present invention relates to target detection technique fields, in particular to a kind of smog detection method, Smoke Detection system
System, computer installation and computer readable storage medium.
Background technology
With the continuous generation of video detecting method, video detection effect also becomes focus.In recent years, video detection
There are many technologies, such as Faster R-CNN (FasterRegion-based Convolutional can be passed through
Neural Networks, fast convolution neural network) and SSD (Single Shot MultiboxDetector, single deep layer
Neural network) model carries out image recognition in video detection.However, existing Faster R-CNN models and SSD models exist
There are still certain defects during video detection, all there is the problems such as precision is low, it is high to calculate cost and time-consuming.
Currently, Smoke Detection belongs to the detection identification problem of specific objective in computer vision field, it is real in the prior art
Smoke Detection algorithm in the use of border includes mainly:Smoke Detection based on colouring information, the Smoke Detection based on movable information
With the Smoke Detection based on wavelet analysis.But it carries out Smoke Detection using colouring information and is easy to be done by Similar color target
It disturbs, all has a significant impact to the accurate testing result of smog using the image-forming condition in movement monitoring region etc., utilize wavelet analysis
Method is often just for the smog of specific modality, it is difficult to meet the application demand of some specific occasions.
Therefore, it is necessary to find a kind of new video detection technology, can in real time be examined in the video data of big data quantity
It measures smog and has become technical problem urgently to be resolved hurrily.
Invention content
The present invention is directed to solve at least one of the technical problems existing in the prior art or related technologies.
For this purpose, one aspect of the present invention is to propose a kind of smog detection method.
Another aspect of the present invention is to propose a kind of smoke detection system.
Another aspect of the invention is to propose a kind of computer installation.
An additional aspect of the present invention is to propose a kind of computer readable storage medium.
In view of this, according to an aspect of the present invention, it proposes a kind of smog detection methods, including:Obtain smog, cigarette
The image data collection of chimney, and to image data collection point of addition label and class label;Location tags and class label will be carried
Image data collection be input in CNN models and be trained as training sample set, obtain the CNN models after multiple training, and
Counting loss function while being trained;Optimal CNN moulds are chosen according in CNN models of the loss function after multiple training
Type carries out Smoke Detection to real-time video, to obtain Smoke Detection result.
Smog detection method provided by the invention obtains the picture of smog and chimney, by smog in picture and chimney first
The position at place is labeled in the form of rectangle frame, i.e. point of addition label, then to the smog and cigarette after point of addition label
The picture of chimney adds class label, and further, the picture of the smog and chimney of point of addition label and class label is carried out
After processing CNN (Convolutional Neural Networks, convolutional neural networks) model is input to as training sample set
In, the CNN models after multiple training are obtained after successive ignition is trained, further, by loss function in multiple training
The best CNN models of training effect, i.e., optimal CNN models, by optimal CNN models to regarding in real time are selected in CNN models afterwards
Frequency carries out Smoke Detection, and provides the testing result that whether there is smog in current real-time video.Smog inspection provided by the invention
Survey method can timely detect smoke by identifying smog and chimney simultaneously in the video of big data quantity in real-time video
Mist can not only promote the accuracy rate of Smoke Detection, additionally it is possible to promote the detection speed of smog.
In addition, CNN models are using Inception-Resnet-V2 (pre-training convolutional neural networks), model training effect is more
Good, loss function uses the optimal CNN model realizations Smoke Detection precision higher that softmax (normalization) function is chosen.
Above-mentioned smog detection method according to the present invention can also have following technical characteristic:
In the above-mentioned technical solutions, it is preferable that obtain the image data collection of smog, chimney, and image data collection is added
It the step of location tags and class label, specifically includes:The image data of smog, chimney is obtained from the video data of collection
Collection, point of addition label and class label on image data concentration smog, chimney location.
In the technical scheme, the video data being collected into is converted to the image data collection of smog, chimney, in picture number
According to concentrate smog and position where chimney in the form of rectangle frame smog, chimney are labeled, i.e. point of addition label,
And class label is added to smog, chimney.By that image data collection point of addition label and class label, can make optimal
Smog and chimney are identified when CNN models carry out Smoke Detection to real-time video simultaneously, and then can promote the speed of Smoke Detection
And accuracy rate.
Wherein, the class label of smog is 0, and the class label of chimney is 1.
In any of the above-described technical solution, it is preferable that using the image data collection with location tags and class label as
Training sample set is input in CNN models and is trained, and obtains the CNN models after multiple training, and while being trained
It the step of counting loss function, specifically includes:By with location tags and class label smog picture with carry location tags
It stores with the chimney picture of class label, and is input to respectively as smog training sample set and chimney training sample set respectively
It is trained in CNN models, respectively obtains the chimney CNN models after the smog CNN models after multiple training and multiple training, and
Fume loss function and stack losses function are calculated separately while being trained;By the smog CNN models after any training
It is added to obtain the CNN models after multiple training with the chimney CNN models after any training, by fume loss function and stack losses
Function is added to obtain loss function;Wherein, training sample set is made of smog training sample set and chimney training sample set.
In the technical scheme, the smog picture with location tags and class label is defeated as smog training sample set
Enter into CNN models and be trained, obtain the smog CNN models after multiple training, smog damage is calculated while being trained
Lose function;Using the chimney picture with location tags and class label as chimney training sample set be input in CNN models into
Row training, obtains the chimney CNN models after multiple training, and stack losses function is calculated while being trained;Further,
By the CNN models for being added smog CNN models with chimney CNN models to obtain after multiple training, by fume loss function and cigarette
Chimney loss function is added to obtain loss function.After smog training sample set and chimney training sample set are trained respectively
The smog CNN models after multiple training and chimney CNN models can be obtained, and then can more accurately identify smog and cigarette
Chimney, to promote the detection result to smog.
In any of the above-described technical solution, it is preferable that chosen most according in CNN models of the loss function after multiple training
Excellent CNN models carry out Smoke Detection to real-time video, the step of to obtain Smoke Detection result, specifically include:Choose loss letter
CNN models after the training of number result of calculation minimum utilize optimal CNN model inspections real-time video as optimal CNN models
The distance that location of smoke moves in front and back frame picture, distance is compared with predetermined threshold value;If distance is more than or equal to default
Threshold value, then it is assumed that detect smog;If distance is less than predetermined threshold value, then it is assumed that do not detect smog.
In the technical scheme, gradient optimal method optimization loss function may be used makes loss most to find best initial weights
It is small, further, according to the CNN models after the training of loss function result of calculation minimum as optimal CNN models to regarding in real time
Frequency carries out Smoke Detection, is compared by the distance and predetermined threshold value that move location of smoke in front and back frame picture in real-time video
Compared with to judge that the real-time video whether there is smog.Smoke Detection is carried out to real-time video by obtained optimal CNN models,
It can timely detect smog, detection speed can also be promoted while promoting the accuracy rate of Smoke Detection.
According to another aspect of the present invention, it is proposed that a kind of smoke detection system, including:Acquiring unit, for obtaining
The image data collection of smog, chimney, and to image data collection point of addition label and class label;Training unit is used for band
There is the image data collection of location tags and class label to be input in CNN models as training sample set to be trained, obtain more
CNN models after a training, and counting loss function while being trained;Detection unit, for being existed according to loss function
Optimal CNN models are chosen in CNN models after multiple training, Smoke Detection is carried out to real-time video, to obtain Smoke Detection knot
Fruit.
Smoke detection system provided by the invention obtains the picture of smog and chimney, by picture by acquiring unit first
Position where middle smog and chimney is labeled in the form of rectangle frame, i.e. point of addition label, then to point of addition label
The picture of smog and chimney afterwards adds class label, further, by the smog and cigarette of point of addition label and class label
The picture of chimney is input to as training sample set in CNN models after being handled, and multiple instructions are obtained after successive ignition is trained
CNN models after white silk, it is further, best by selecting training effect in CNN models of the loss function after multiple training
CNN models, i.e., optimal CNN models carry out Smoke Detection to real-time video by optimal CNN models, and provide and currently regard in real time
It whether there is the testing result of smog in frequency.Smog detection method provided by the invention by the video of big data quantity simultaneously
It identifies smog and chimney, smog can be timely detected in real-time video, the accuracy rate of Smoke Detection can not only be promoted,
The detection speed of smog can also be promoted.
In addition, CNN models are more preferable using Inception-Resnet-V2 model training effects, loss function uses
The optimal CNN model realizations Smoke Detection precision higher that softmax functions are chosen.
The control system of above-mentioned mobile terminal according to the present invention can also have following technical characteristic:
In the above-mentioned technical solutions, it is preferable that acquiring unit is specifically used for:From the video data of collection obtain smog,
The image data collection of chimney, point of addition label and class label on image data concentration smog, chimney location.
In the technical scheme, the video data being collected into is converted to the image data collection of smog, chimney, in picture number
According to concentrate smog and position where chimney in the form of rectangle frame smog, chimney are labeled, i.e. point of addition label,
And class label is added to smog, chimney.By that image data collection point of addition label and class label, can make optimal
Smog and chimney are identified when CNN models carry out Smoke Detection to real-time video simultaneously, and then can promote the speed of Smoke Detection
And accuracy rate.
Wherein, the class label of smog is 0, and the class label of chimney is 1.
In any of the above-described technical solution, it is preferable that training unit specifically includes:Computing unit, for position will to be carried
The smog picture of label and class label stores respectively with the chimney picture with location tags and class label, and respectively as
Smog training sample set and chimney training sample set, which are input in CNN models, to be trained, and the cigarette after multiple training is respectively obtained
Chimney CNN models after mist CNN models and multiple training, and fume loss function and cigarette are calculated separately while being trained
Chimney loss function;Summation unit is used for the smog CNN models after any training and the chimney CNN model phases after any training
Add to obtain the CNN models after multiple training, fume loss function is added to obtain loss function with stack losses function;Wherein,
Training sample set is made of smog training sample set and chimney training sample set.
In the technical scheme, the smog picture with location tags and class label is defeated as smog training sample set
Enter into CNN models and be trained, obtains the smog CNN models after multiple training, it is single by calculating while being trained
Member calculates fume loss function;Chimney picture with location tags and class label is input to as chimney training sample set
It is trained in CNN models, obtains the chimney CNN models after multiple training, pass through computing unit meter while being trained
Calculate stack losses function;Further, smog CNN models are added to obtain multiple instructions with chimney CNN models by summation unit
Fume loss function is added to obtain loss function by the CNN models after white silk with stack losses function.By by smog training sample
Collection and chimney training sample set can obtain the smog CNN models after multiple training and chimney CNN models after being trained respectively,
And then can more accurately identify smog and chimney, to promote the detection result to smog.
In any of the above-described technical solution, it is preferable that detection unit is specifically used for:It is minimum to choose loss function result of calculation
Training after model as optimal CNN models, and utilize smog position in frame picture before and after optimal CNN model inspections real-time video
Mobile distance is set, distance is compared with predetermined threshold value;If distance is more than or equal to predetermined threshold value, then it is assumed that detect cigarette
Mist;If distance is less than predetermined threshold value, then it is assumed that do not detect smog.
In the technical scheme, gradient optimal method optimization loss function may be used makes loss most to find best initial weights
It is small, further, according to the CNN models after the training of loss function result of calculation minimum as optimal CNN models to regarding in real time
Frequency carries out Smoke Detection, is compared by the distance and predetermined threshold value that move location of smoke in front and back frame picture in real-time video
Compared with to judge that the real-time video whether there is smog.Smoke Detection is carried out to real-time video by obtained optimal CNN models,
It can timely detect smog, detection speed can also be promoted while promoting the accuracy rate of Smoke Detection.
According to a further aspect of the invention, it is proposed that a kind of computer installation, including memory, processor and be stored in
On memory and the computer program that can run on a processor, processor realize such as any of the above-described when executing computer program
In smog detection method.
Computer installation provided by the invention is realized when executing computer program by processor as in any of the above-described
Smog detection method realizes the image data collection for obtaining smog, chimney, and to image data collection point of addition label and classification mark
Label;Image data collection with location tags and class label is input to as training sample set in CNN models and is trained,
Obtain the CNN models after multiple training, and counting loss function while being trained;According to loss function in multiple training
Optimal CNN models are chosen in CNN models afterwards, Smoke Detection is carried out to real-time video, to obtain Smoke Detection result.
According to a further aspect of the invention, it is proposed that a kind of computer readable storage medium is stored thereon with computer
Program is realized when computer program is executed by processor such as the smog detection method in any of the above-described.
Computer readable storage medium provided by the invention is realized when being executed by processor by computer program as above-mentioned
Smog detection method in any one realizes the image data collection for obtaining smog, chimney, and to image data collection point of addition mark
Label and class label;Image data collection with location tags and class label is input to CNN models as training sample set
In be trained, obtain the CNN models after multiple training, and counting loss function while being trained;According to loss letter
Number chooses optimal CNN models and carries out Smoke Detection to real-time video in the CNN models after multiple training, to obtain Smoke Detection
As a result.
The additional aspect and advantage of the present invention will become apparent in following description section, or practice through the invention
Recognize.
Description of the drawings
The above-mentioned and/or additional aspect and advantage of the present invention will become in the description from combination following accompanying drawings to embodiment
Obviously and it is readily appreciated that, wherein:
Fig. 1 shows the flow diagram of the smog detection method of one embodiment of the present of invention;
Fig. 2 shows the flow diagrams of the smog detection method of an alternative embodiment of the invention;
Fig. 3 shows the flow diagram of the smog detection method of yet another embodiment of the present invention;
Fig. 4 shows the flow diagram of the smog detection method of another embodiment of the present invention;
Fig. 5 a show the schematic block diagram of the smoke detection system of one embodiment of the present of invention;
Figure 5b shows that the schematic block diagrams of the smoke detection system of an alternative embodiment of the invention;
Fig. 6 shows the schematic block diagram of the computer installation of one embodiment of the present of invention.
Specific implementation mode
To better understand the objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and specific real
Mode is applied the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
Many details are elaborated in the following description to facilitate a thorough understanding of the present invention, still, the present invention may be used also
To be implemented different from other modes described here using other, therefore, protection scope of the present invention is not by described below
Specific embodiment limitation.
Referring to the smog detection method described in Fig. 1 to Fig. 6 descriptions according to some embodiments of the invention, Smoke Detection system
System, computer installation and computer readable storage medium.
The embodiment of first aspect present invention proposes that a kind of smog detection method, Fig. 1 show the implementation of the present invention
The flow diagram of the smog detection method of example, wherein this approach includes the following steps:
Step S102 obtains the image data collection of smog, chimney, and to image data collection point of addition label and classification mark
Label;
Image data collection with location tags and class label is input to CNN by step S104 as training sample set
It is trained in model, obtains the CNN models after multiple training, and counting loss function while being trained;
Step S106 chooses optimal CNN models to real-time video according in CNN models of the loss function after multiple training
Smoke Detection is carried out, to obtain Smoke Detection result.
Smog detection method provided by the invention obtains the picture of smog and chimney, by smog in picture and chimney first
The position at place is labeled in the form of rectangle frame, i.e. point of addition label, then to the smog and cigarette after point of addition label
The picture of chimney adds class label, and further, the picture of the smog and chimney of point of addition label and class label is carried out
It is input in CNN models as training sample set after processing, the CNN moulds after multiple training is obtained after successive ignition is trained
Type, further, by selecting the best CNN models of training effect in CNN models of the loss function after multiple training, i.e., most
Excellent CNN models carry out Smoke Detection to real-time video by optimal CNN models, and provide and whether there is in current real-time video
The testing result of smog.Smog detection method provided by the invention by identifying smog and cigarette simultaneously in the video of big data quantity
Chimney can timely detect smog in real-time video, can not only promote the accuracy rate of Smoke Detection, additionally it is possible to promote cigarette
The detection speed of mist.
In addition, CNN models are more preferable using Inception-Resnet-V2 model training effects, loss function uses
The optimal CNN model realizations Smoke Detection precision higher that softmax functions are chosen.
In one embodiment of the invention, Fig. 2 shows the smog detection methods of an alternative embodiment of the invention
Flow diagram, wherein this approach includes the following steps:
Step S202 obtains the image data collection of smog, chimney from the video data of collection, and cigarette is concentrated in image data
Point of addition label and class label on mist, chimney location;
Image data collection with location tags and class label is input to CNN by step S204 as training sample set
It is trained in model, obtains the CNN models after multiple training, and counting loss function while being trained;
Step S206 chooses optimal CNN models to real-time video according in CNN models of the loss function after multiple training
Smoke Detection is carried out, to obtain Smoke Detection result.
In this embodiment, the video data being collected into is converted to the image data collection of smog, chimney, in image data
Smog, chimney are labeled in the form of rectangle frame at position where concentration smog and chimney, i.e. point of addition label, and
Class label is added to smog, chimney.By the way that image data collection point of addition label and class label, optimal CNN can be made
Smog and chimney are identified when model carries out Smoke Detection to real-time video simultaneously, and then can promote the speed and standard of Smoke Detection
True rate.
Wherein, the class label of smog is 0, and the class label of chimney is 1.
In one embodiment of the invention, Fig. 3 shows the smog detection method of yet another embodiment of the present invention
Flow diagram, wherein this approach includes the following steps:
Step S302 obtains the image data collection of smog, chimney from the video data of collection, and cigarette is concentrated in image data
Point of addition label and class label on mist, chimney location;
Step S304, by with location tags and class label smog picture with location tags and class label
Chimney picture stores respectively, and is input in CNN models and carries out respectively as smog training sample set and chimney training sample set
Training, respectively obtain the chimney CNN models after the smog CNN models after multiple training and multiple training, and is being trained
Calculate separately fume loss function and stack losses function simultaneously;
Smog CNN models after any training with the chimney CNN models after any training are added to obtain more by step S306
Fume loss function is added to obtain loss function by the CNN models after a training with stack losses function;
Step S308 chooses optimal CNN models to real-time video according in CNN models of the loss function after multiple training
Smoke Detection is carried out, to obtain Smoke Detection result.
In this embodiment, it is inputted the smog picture with location tags and class label as smog training sample set
To being trained in CNN models, the smog CNN models after multiple training are obtained, fume loss is calculated while being trained
Function;Chimney picture with location tags and class label is input to as chimney training sample set in CNN models and is carried out
Training, obtains the chimney CNN models after multiple training, and stack losses function is calculated while being trained;Further, lead to
It crosses smog CNN models and to be added to obtain the CNN models after multiple training with chimney CNN models, by fume loss function and chimney
Loss function is added to obtain loss function.Pass through energy after being trained smog training sample set and chimney training sample set respectively
The smog CNN models after multiple training and chimney CNN models are accessed, and then can more accurately identify smog and chimney,
To promote the detection result to smog.
In one embodiment of the invention, Fig. 4 shows the smog detection method of another embodiment of the present invention
Flow diagram, wherein this approach includes the following steps:
Step S402 obtains the image data collection of smog, chimney from the video data of collection, and cigarette is concentrated in image data
Point of addition label and class label on mist, chimney location;
Step S404, by with location tags and class label smog picture with location tags and class label
Chimney picture stores respectively, and is input in CNN models and carries out respectively as smog training sample set and chimney training sample set
Training, respectively obtain the chimney CNN models after the smog CNN models after multiple training and multiple training, and is being trained
Calculate separately fume loss function and stack losses function simultaneously;
Smog CNN models after any training with the chimney CNN models after any training are added to obtain more by step S406
Fume loss function is added to obtain loss function by the CNN models after a training with stack losses function;
Step S408 chooses the CNN models after the training of loss function result of calculation minimum as optimal CNN models, and
The distance moved using location of smoke in frame picture before and after optimal CNN model inspections real-time video;
Step S410, judges whether the distance of location of smoke movement is more than or equal to predetermined threshold value;
Step S412, if distance is more than or equal to predetermined threshold value, then it is assumed that detect smog;
Step S414, if distance is less than predetermined threshold value, then it is assumed that do not detect smog.
In this embodiment it is possible to which gradient optimal method optimization loss function is used to make loss most to find best initial weights
It is small, further, according to the CNN models after the training of loss function result of calculation minimum as optimal CNN models to regarding in real time
Frequency carries out Smoke Detection, is compared by the distance and predetermined threshold value that move location of smoke in front and back frame picture in real-time video
Compared with to judge that the real-time video whether there is smog.Smoke Detection is carried out to real-time video by obtained optimal CNN models,
It can timely detect smog, detection speed can also be promoted while promoting the accuracy rate of Smoke Detection.
The embodiment of second aspect of the present invention proposes that a kind of smoke detection system, Fig. 5 a show the reality of the present invention
Apply the schematic block diagram of the smoke detection system 500 of example, wherein the system includes:
Acquiring unit 502, the image data collection for obtaining smog, chimney, and to image data collection point of addition label
And class label;
Training unit 504, the image data collection for that will carry location tags and class label are defeated as training sample set
Enter into CNN models and be trained, obtains the CNN models after multiple training, and counting loss function while being trained;
Detection unit 506, for choosing optimal CNN models pair according in CNN models of the loss function after multiple training
Real-time video carries out Smoke Detection, to obtain Smoke Detection result.
Smoke detection system 500 provided by the invention obtains the picture of smog and chimney by acquiring unit 502 first,
Smog in picture and the position where chimney are labeled in the form of rectangle frame, i.e. point of addition label, then to adding position
The picture addition class label for setting the smog after label and chimney, further, by the cigarette of point of addition label and class label
The picture of mist and chimney is input to as training sample set in CNN models after being handled, and is obtained after successive ignition is trained
CNN models after multiple training, further, by selecting training effect in CNN models of the loss function after multiple training
Best CNN models, i.e., optimal CNN models carry out Smoke Detection to real-time video by optimal CNN models, and provide current
It whether there is the testing result of smog in real-time video.Smog detection method provided by the invention passes through the video in big data quantity
In simultaneously identify smog and chimney, smog can be timely detected in real-time video, Smoke Detection can not only be promoted
Accuracy rate, additionally it is possible to promote the detection speed of smog.
In addition, CNN models are more preferable using Inception-Resnet-V2 model training effects, loss function uses
The optimal CNN model realizations Smoke Detection precision higher that softmax functions are chosen.
In one embodiment of the invention, it is preferable that acquiring unit 502 is specifically used for:From the video data of collection
The image data collection for obtaining smog, chimney, point of addition label and class on image data concentration smog, chimney location
Distinguishing label.
In this embodiment, the video data being collected into is converted to the image data collection of smog, chimney, in image data
Smog, chimney are labeled in the form of rectangle frame at position where concentration smog and chimney, i.e. point of addition label, and
Class label is added to smog, chimney.By the way that image data collection point of addition label and class label, optimal CNN can be made
Smog and chimney are identified when model carries out Smoke Detection to real-time video simultaneously, and then can promote the speed and standard of Smoke Detection
True rate.
Wherein, the class label of smog is 0, and the class label of chimney is 1.
In one embodiment of the invention, it is preferable that Figure 5b shows that the inspections of the smog of an alternative embodiment of the invention
The schematic block diagram of examining system 500, wherein the system includes:
Acquiring unit 502, the image data collection for obtaining smog, chimney from the video data of collection, in picture number
According to point of addition label and class label on concentration smog, chimney location;
Training unit 504, the image data collection for that will carry location tags and class label are defeated as training sample set
Enter into CNN models and be trained, obtains the CNN models after multiple training, and counting loss function while being trained;
Training unit 504 includes computing unit 508, and the smog picture for that will carry location tags and class label is marked with position
The chimney picture of label and class label stores respectively, and is input to respectively as smog training sample set and chimney training sample set
It is trained in CNN models, respectively obtains the chimney CNN models after the smog CNN models after multiple training and multiple training, and
Fume loss function and stack losses function are calculated separately while being trained;Summation unit 510 is used for any training
Smog CNN models afterwards are added to obtain the CNN models after multiple training with the chimney CNN models after any training, and smog is damaged
Function is lost to be added to obtain loss function with stack losses function;Wherein, training sample set is instructed by smog training sample set and chimney
Practice sample set composition.
Detection unit 506, for choosing optimal CNN models pair according in CNN models of the loss function after multiple training
Real-time video carries out Smoke Detection, to obtain Smoke Detection result.
In this embodiment, it is inputted the smog picture with location tags and class label as smog training sample set
To being trained in CNN models, the smog CNN models after multiple training are obtained, pass through computing unit while being trained
508 calculate fume loss function;It is inputted the chimney picture with location tags and class label as chimney training sample set
To being trained in CNN models, the chimney CNN models after multiple training are obtained, pass through computing unit while being trained
508 calculate stack losses function;Further, smog CNN models are added with chimney CNN models by summation unit 510
Fume loss function is added to obtain loss function by the CNN models after to multiple training with stack losses function.By by smog
Training sample set and chimney training sample set can obtain the smog CNN models and chimney after multiple training after being trained respectively
CNN models, and then can more accurately identify smog and chimney, to promote the detection result to smog.
In one embodiment of the invention, it is preferable that detection unit 506 is specifically used for:It chooses loss function and calculates knot
Model after the training of fruit minimum is as optimal CNN models, and using in frame picture before and after optimal CNN model inspections real-time video
The distance of location of smoke movement, distance is compared with predetermined threshold value;If distance is more than or equal to predetermined threshold value, then it is assumed that inspection
Measure smog;If distance is less than predetermined threshold value, then it is assumed that do not detect smog.
In this embodiment it is possible to which gradient optimal method optimization loss function is used to make loss most to find best initial weights
It is small, further, according to the CNN models after the training of loss function result of calculation minimum as optimal CNN models to regarding in real time
Frequency carries out Smoke Detection, is compared by the distance and predetermined threshold value that move location of smoke in front and back frame picture in real-time video
Compared with to judge that the real-time video whether there is smog.Smoke Detection is carried out to real-time video by obtained optimal CNN models,
It can timely detect smog, detection speed can also be promoted while promoting the accuracy rate of Smoke Detection.
The embodiment of third aspect present invention proposes that a kind of computer installation, Fig. 6 show one embodiment of the present of invention
Computer installation 600 schematic block diagram, wherein the device includes:Memory 602, processor 604 and it is stored in memory
On 602 and the computer program that can be run on processor 604, processor 604 are realized when executing computer program as above-mentioned
Smog detection method.
Computer installation 600 provided by the invention, processor 604 are realized when executing computer program by big data quantity
Video in simultaneously identify smog and chimney, smog can be timely detected in real-time video, smog can not only be promoted
The accuracy rate of detection, additionally it is possible to promote the detection speed of smog.
The embodiment of fourth aspect present invention proposes a kind of computer readable storage medium, is stored thereon with computer journey
Sequence realizes such as above-mentioned smog detection method when computer program is executed by processor 604.
Computer readable storage medium provided by the invention, is stored thereon with computer program, and computer program is handled
Realized when device 604 executes such as above-mentioned smog detection method, realize by identified simultaneously in the video of big data quantity smog with
Chimney can timely detect smog in real-time video, can not only promote the accuracy rate of Smoke Detection, additionally it is possible to be promoted
The detection speed of smog.
In the present invention, term " multiple " then refers to two or more, unless otherwise restricted clearly.Term " peace
The terms such as dress ", " connected ", " connection ", " fixation " shall be understood in a broad sense, for example, " connection " may be a fixed connection, it can also
It is to be detachably connected, or be integrally connected;" connected " can be directly connected, can also be indirectly connected through an intermediary.It is right
For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific conditions.
In the description of this specification, the description of term " one embodiment ", " some embodiments ", " specific embodiment " etc.
Mean that particular features, structures, materials, or characteristics described in conjunction with this embodiment or example are contained at least one reality of the present invention
It applies in example or example.In the present specification, schematic expression of the above terms are not necessarily referring to identical embodiment or reality
Example.Moreover, description particular features, structures, materials, or characteristics can in any one or more of the embodiments or examples with
Suitable mode combines.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, any made by repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of smog detection method, which is characterized in that including:
The image data collection of smog, chimney is obtained, and to the image data collection point of addition label and class label;
The image data collection with location tags and class label is input to as training sample set in CNN models and is carried out
Training, obtains the CNN models after multiple training, and counting loss function while being trained;
Real-time video is carried out according to optimal CNN models are chosen in CNN models of the loss function after the multiple training
Smoke Detection, to obtain Smoke Detection result.
2. smog detection method according to claim 1, which is characterized in that the image data for obtaining smog, chimney
Collection, and the step of to the image data collection point of addition label and class label, specifically include:
The image data collection that the smog, chimney are obtained from the video data of collection concentrates smog, cigarette in the image data
Point of addition label and class label on chimney location.
3. smog detection method according to claim 1 or 2, which is characterized in that described to carry location tags and classification
The image data collection of label, which is input to as training sample set in CNN models, to be trained, and the CNN after multiple training is obtained
Model, and specifically included the step of counting loss function while being trained:
Smog picture with location tags and class label is distinguished with the chimney picture with location tags and class label
Storage, and be input in CNN models and be trained respectively as smog training sample set and chimney training sample set, it respectively obtains
The chimney CNN models after smog CNN models and multiple training after multiple training, and cigarette is calculated separately while being trained
Mist loss function and stack losses function;
Smog CNN models after any training are added with the chimney CNN models after any training after obtaining the multiple training
CNN models, the fume loss function is added to obtain the loss function with the stack losses function;
Wherein, the training sample set is made of the smog training sample set and the chimney training sample set.
4. smog detection method according to claim 3, which is characterized in that it is described according to the loss function described more
Optimal CNN models are chosen in CNN models after a training, Smoke Detection is carried out to real-time video, to obtain Smoke Detection result
Step specifically includes:
The CNN models after the training of the loss function result of calculation minimum are chosen as the optimal CNN models, and utilize institute
The distance that location of smoke moves in frame picture before and after optimal CNN model inspections real-time video is stated, by the distance and predetermined threshold value
It is compared;
If the distance is more than or equal to the predetermined threshold value, then it is assumed that detect smog;
If the distance is less than the predetermined threshold value, then it is assumed that do not detect smog.
5. a kind of smoke detection system, which is characterized in that including:
Acquiring unit, the image data collection for obtaining smog, chimney, and to the image data collection point of addition label and class
Distinguishing label;
Training unit, for the image data collection for carrying location tags and class label to be input to as training sample set
It is trained in CNN models, obtains the CNN models after multiple training, and counting loss function while being trained;
Detection unit, for choosing optimal CNN models pair according in CNN models of the loss function after the multiple training
Real-time video carries out Smoke Detection, to obtain Smoke Detection result.
6. smoke detection system according to claim 5, which is characterized in that the acquiring unit is specifically used for:
The image data collection that the smog, chimney are obtained from the video data of collection concentrates smog, cigarette in the image data
Point of addition label and class label on chimney location.
7. smoke detection system according to claim 5 or 6, which is characterized in that the training unit specifically includes:
Computing unit, for by with location tags and class label smog picture with location tags and class label
Chimney picture stores respectively, and is input in CNN models and carries out respectively as smog training sample set and chimney training sample set
Training, respectively obtain the chimney CNN models after the smog CNN models after multiple training and multiple training, and is being trained
Calculate separately fume loss function and stack losses function simultaneously;
Summation unit, for the smog CNN models after any training to be added to obtain institute with the chimney CNN models after any training
The CNN models after multiple training are stated, the fume loss function is added with the stack losses function to obtain the loss letter
Number;
Wherein, the training sample set is made of the smog training sample set and the chimney training sample set.
8. smoke detection system according to claim 7, which is characterized in that the detection unit is specifically used for:
The model after the training of the loss function result of calculation minimum is chosen as the optimal CNN models, and described in utilization
The distance that location of smoke moves in frame picture before and after optimal CNN model inspections real-time video, by the distance and predetermined threshold value into
Row compares;
If the distance is more than or equal to the predetermined threshold value, then it is assumed that detect smog;
If the distance is less than the predetermined threshold value, then it is assumed that do not detect smog.
9. a kind of computer installation, including memory, processor and it is stored on the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as Claims 1-4 when executing the computer program
Any one of described in smog detection method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
Smog detection method according to any one of claims 1 to 4 is realized when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810217094.6A CN108764264A (en) | 2018-03-16 | 2018-03-16 | Smog detection method, smoke detection system and computer installation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810217094.6A CN108764264A (en) | 2018-03-16 | 2018-03-16 | Smog detection method, smoke detection system and computer installation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108764264A true CN108764264A (en) | 2018-11-06 |
Family
ID=63980406
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810217094.6A Pending CN108764264A (en) | 2018-03-16 | 2018-03-16 | Smog detection method, smoke detection system and computer installation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108764264A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109766780A (en) * | 2018-12-20 | 2019-05-17 | 武汉理工大学 | A kind of ship smog emission on-line checking and method for tracing based on deep learning |
CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
CN112215122A (en) * | 2020-09-30 | 2021-01-12 | 中国科学院深圳先进技术研究院 | Fire detection method, system, terminal and storage medium based on video image target detection |
CN112634151A (en) * | 2020-12-14 | 2021-04-09 | 深圳中兴网信科技有限公司 | Poisson fusion-based smoke data enhancement method, enhancement equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105741479A (en) * | 2016-01-25 | 2016-07-06 | 赣州市金电电子设备有限公司 | Integrated forest fire prevention IA-PCNN algorithm based on thermal imaging and smoke identification |
CN106778820A (en) * | 2016-11-25 | 2017-05-31 | 北京小米移动软件有限公司 | Identification model determines method and device |
CN107025443A (en) * | 2017-04-06 | 2017-08-08 | 江南大学 | Stockyard smoke monitoring and on-time model update method based on depth convolutional neural networks |
-
2018
- 2018-03-16 CN CN201810217094.6A patent/CN108764264A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105741479A (en) * | 2016-01-25 | 2016-07-06 | 赣州市金电电子设备有限公司 | Integrated forest fire prevention IA-PCNN algorithm based on thermal imaging and smoke identification |
CN106778820A (en) * | 2016-11-25 | 2017-05-31 | 北京小米移动软件有限公司 | Identification model determines method and device |
CN107025443A (en) * | 2017-04-06 | 2017-08-08 | 江南大学 | Stockyard smoke monitoring and on-time model update method based on depth convolutional neural networks |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109766780A (en) * | 2018-12-20 | 2019-05-17 | 武汉理工大学 | A kind of ship smog emission on-line checking and method for tracing based on deep learning |
CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
CN112215122A (en) * | 2020-09-30 | 2021-01-12 | 中国科学院深圳先进技术研究院 | Fire detection method, system, terminal and storage medium based on video image target detection |
CN112215122B (en) * | 2020-09-30 | 2023-10-24 | 中国科学院深圳先进技术研究院 | Fire detection method, system, terminal and storage medium based on video image target detection |
CN112634151A (en) * | 2020-12-14 | 2021-04-09 | 深圳中兴网信科技有限公司 | Poisson fusion-based smoke data enhancement method, enhancement equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103324937B (en) | The method and apparatus of label target | |
CN108764264A (en) | Smog detection method, smoke detection system and computer installation | |
CN111242123B (en) | Power equipment fault diagnosis method based on infrared image | |
CN103279765B (en) | Steel wire rope surface damage detection method based on images match | |
CN105956632B (en) | A kind of method and apparatus detecting target | |
CN109285139A (en) | A kind of x-ray imaging weld inspection method based on deep learning | |
CN108810620A (en) | Identify method, computer equipment and the storage medium of the material time point in video | |
CN110288032B (en) | Vehicle driving track type detection method and device | |
CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
CN109344905A (en) | A kind of transmission facility automatic fault recognition methods based on integrated study | |
CN109360183A (en) | A kind of quality of human face image appraisal procedure and system based on convolutional neural networks | |
CN108509976A (en) | The identification device and method of animal | |
CN103325122B (en) | Based on the pedestrian retrieval method of Bidirectional sort | |
CN104615986B (en) | The method that pedestrian detection is carried out to the video image of scene changes using multi-detector | |
CN113361354B (en) | Track component inspection method and device, computer equipment and storage medium | |
CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
CN109117836A (en) | Text detection localization method and device under a kind of natural scene based on focal loss function | |
CN110399884A (en) | A kind of adaptive anchor frame model vehicle checking method of Fusion Features | |
CN110378232B (en) | Improved test room examinee position rapid detection method of SSD dual-network | |
CN103871077B (en) | A kind of extraction method of key frame in road vehicles monitoring video | |
CN110276264A (en) | A kind of crowd density estimation method based on foreground segmentation figure | |
CN109886947A (en) | The high-tension bus-bar defect inspection method of convolutional neural networks based on region | |
CN105866782B (en) | A kind of moving object detection system and method based on laser radar | |
CN108960142B (en) | Pedestrian re-identification method based on global feature loss function | |
CN101930533A (en) | Device and method for performing sky detection in image collecting device |
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: 20181106 |