CN110070106A - Smog detection method, device and electronic equipment - Google Patents
Smog detection method, device and electronic equipment Download PDFInfo
- Publication number
- CN110070106A CN110070106A CN201910236445.2A CN201910236445A CN110070106A CN 110070106 A CN110070106 A CN 110070106A CN 201910236445 A CN201910236445 A CN 201910236445A CN 110070106 A CN110070106 A CN 110070106A
- Authority
- CN
- China
- Prior art keywords
- video
- classification
- classification results
- processing
- obtains
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 166
- 238000012545 processing Methods 0.000 claims abstract description 134
- 239000000779 smoke Substances 0.000 claims abstract description 73
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 70
- 238000000034 method Methods 0.000 claims description 84
- 230000006870 function Effects 0.000 claims description 48
- 238000013473 artificial intelligence Methods 0.000 claims description 32
- 239000013598 vector Substances 0.000 claims description 24
- 239000003595 mist Substances 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- 235000013399 edible fruits Nutrition 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 description 36
- 238000012549 training Methods 0.000 description 18
- 238000010586 diagram Methods 0.000 description 15
- 239000011159 matrix material Substances 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 10
- 235000019504 cigarettes Nutrition 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000011897 real-time detection Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 108010001267 Protein Subunits Proteins 0.000 description 1
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
- G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Emergency Management (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Business, Economics & Management (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application provides a kind of smog detection method, device and electronic equipment, is related to field of image detection, comprising: obtains Smoke Detection video;Smoke Detection video is pre-processed, obtains processing video, processing video includes multiple processing frames;Single frames sort operation is carried out to each processing frame according to the first intelligent algorithm, obtains the first classification results;When the first classification results include for indicating the first judging result existing for smog, multiframe sort operation is carried out to multiple processing frames according to the second intelligent algorithm, obtains the second classification results;When the second classification results include the first judging result, the first judging result is exported.Implement this embodiment, can be improved the precision of Smoke Detection, the rate that reports an error of Smoke Detection is reduced, to improve wide usage.
Description
Technical field
This application involves technical field of image detection, set in particular to a kind of smog detection method, device and electronics
It is standby.
Background technique
Currently, the prior art would generally detect smoke region and be classified using the method for machine learning, be because
Smoke Detection and classification can be carried out to video sequence by the feature of Manual definition for such method, and there is arithmetic speed
Fast and high detection efficiency feature.However, it has been found in practice that above by the method for Manual definition's feature to Smoke Detection
Precision is relatively low, and can also have the case where a large amount of wrong reports using this kind of method, this allows for current smog detection method
It is difficult to be applied in complex scene, lacks wide usage.
Summary of the invention
The embodiment of the present application is designed to provide a kind of smog detection method, device and electronic equipment, to solve cigarette
The precision of mist detection is low, and Smoke Detection has a large amount of wrong reports, the low problem of wide usage.
The embodiment of the present application provides a kind of smog detection method, comprising:
Obtain Smoke Detection video;
The Smoke Detection video is pre-processed, obtains processing video, the processing video includes multiple processing frames;
Single frames sort operation is carried out to each processing frame according to the first intelligent algorithm, obtains the first classification results;
When first classification results include for indicating the first judging result existing for smog, according to the second artificial intelligence
Energy algorithm carries out multiframe sort operation to the multiple processing frame, obtains the second classification results;
When second classification results include first judging result, first judging result is exported.
During above-mentioned realization, video sequence to be detected can be preferentially obtained, and complete the pre- place to video sequence
Reason obtains processing video;After obtaining processing video, single frames classification is carried out to processing video according to the first intelligent algorithm
Processing carries out at multiframe classification processing video according further to the second intelligent algorithm with completing preliminary smog judgement
Reason, to complete the judgement of accurate smog, to obtain higher the first judging result of smog of precision, and last output should as a result,
So that relevant operator learns testing result.As it can be seen that implementing this embodiment, Smoke Detection can be effectively improved
Detection accuracy, and the probability to report an error can be substantially reduced by secondary detection, so that the smog detection method is being practiced
In have wider use.
Further, described that the Smoke Detection video is pre-processed, obtain processing video the step of include:
According to pre-set dimension each image frame dimension adjustment ongoing to the Smoke Detection video, after being adjusted
Adjust video;
The extraction of video frame is carried out to the adjustment video according to default detection mode, and is with the video frame of extraction
According to acquisition motion detection video;
Judge in the motion detection video with the presence or absence of moving object;
If there are when the moving object in the motion detection video, to the movement in the motion detection video
The contour edge of object extracts, and obtains processing video.
During above-mentioned realization, preferentially according to picture size to detection video carry out picture adjustment pretreatment, obtain with
The detection video of desired picture same size, and moving region detection is carried out to the obtained detection video, obtain movement inspection
Survey video;Judge whether the motion detection video is true according to motion detection video, and the situation true in motion detection video
Under (there are in the case where moving object), to the motion detection video carry out moving object contour edge extraction process, obtain
Handle video.Implement this embodiment, pretreated process can be refined, so that the content of the processing video
It is more accurate, so that subsequent detection process is more smooth because of unified specification, can also avoid because of format, size
Disunity and caused by precision problem.
Further, corresponding first artificial intelligence model of first intelligent algorithm includes five convolutional layers, the
One full articulamentum and the first classification layer;Wherein, described that single frames point is carried out to each processing frame according to the first intelligent algorithm
Class operation, the step of obtaining the first classification results include:
Convolution algorithm is carried out to each processing frame according to five convolutional layers, obtains the first characteristics of image group;
Data processing is carried out to the first image feature group according to the described first full articulamentum, obtains first eigenvector
Group;
Classified according to the first classification layer to the first eigenvector group, obtains first classification results.
During above-mentioned realization, the internal structure of the first intelligent algorithm is defined, and specifically define first
Intelligent algorithm is how video and to obtain the first classification results to processing.As it can be seen that implementing this embodiment, Ke Yixi
Change and processing video is handled to obtain the detailed process of the first classification results, and can also determine the first artificial intelligence model
Specific structure so that processing video can obtain accurate first classification results with this configuration, and then improve first
The classification accuracy of classification results;Meanwhile the process is to carry out single frame detection, that is to say, that it detects model in this process
It encloses more greatly, effect is more preferable, and accuracy is also higher.
Further, the first classification layer includes first-loss function;Wherein, described according to the first classification layer
After the step of classifying to the first eigenvector group, obtain the first classification results, the method also includes:
First classification results are calculated according to the first-loss function, obtain the first difference value;
Point that the first classification layer includes is updated according to first difference value and preset stochastic gradient descent algorithm
Class parameter.
During above-mentioned realization, the first classification layer includes first-loss function and sorting parameter in this method, this
One loss function fixes, and for calculating respective value, and adjusting includes sorting parameter (classification ginseng in the first classification layer
It is several with first-loss function and uncorrelated), wherein the sorting parameter is used to improve the first nicety of grading for classifying layer.As it can be seen that can
With by the step similar with training process apply with it is actually detected in, realize actually detected result negative-feedback, so as to
So that the model is able to carry out autonomous iteration, to independently optimize its nicety of grading.
Further, corresponding second artificial intelligence model of second intelligent algorithm includes three convolutional layers, the
Two full articulamentums and the second classification layer;Wherein, described more to the progress of the multiple processing frame according to the second intelligent algorithm
Frame classification operation, the step of obtaining the second classification results include:
According to three convolutional layers convolution algorithm is carried out to the multiple processing frame respectively, obtains the second characteristics of image
Group;
Data processing is carried out to the second characteristics of image group according to the described second full articulamentum, obtains second feature vector
Group;
Classified according to the second classification layer to the second feature Vector Groups, obtains second classification results.
During above-mentioned realization, the internal structure of the second intelligent algorithm is defined, and specifically define second
Intelligent algorithm is how video and to obtain the second classification results to processing.As it can be seen that implementing this embodiment, Ke Yixi
Change and processing video is handled to obtain the detailed process of the second classification results, and can also determine the second artificial intelligence model
Specific structure so that processing video can obtain accurate second classification results with this configuration, and then improve second
The classification accuracy of classification results;Meanwhile the process is to carry out multi frame detection, that is to say, that its detection essence in this process
Du Genggao, effect is more preferable, effectively can carry out high-precision detection to smog;Also, the process can be used as secondary detection
In the presence of so that there are high-precision judgement is realized when smog in the testing result obtained after one-time detection, to realize
The raising for the accuracy that Smoke Detection integrally judges so that this kind of high-precision method can be useful in various environment it
Under, improve its universality.
Further, the second classification layer includes the second loss function;Wherein, described according to the second classification layer
After the step of classifying to the second feature Vector Groups, obtain the second classification results, the method also includes:
Second classification results are calculated according to second loss function, obtain the second difference value;
Point that the second classification layer includes is updated according to second difference value and preset stochastic gradient descent algorithm
Class parameter.
During above-mentioned realization, the second classification layer includes the second loss function and sorting parameter in this method, this
Two loss functions fix, and for calculating respective value, and adjust the sorting parameter (classification for including in the second classification layer
Parameter and the second loss function are simultaneously uncorrelated), wherein the sorting parameter is used to improve the second nicety of grading for classifying layer.As it can be seen that
Implement this embodiment, can by the step similar with training process apply with it is actually detected in, realize actually detected
As a result negative-feedback, so that the model is able to carry out autonomous iteration, to independently optimize its nicety of grading.
Further, the method also includes:
When second classification results do not include first judging result, export for indicating that the second of non smoke sentences
Disconnected result.
During above-mentioned realization, judge also to issue and cigarette is not present there is no when smog in the second classification results
The prompt information of mist, so that testing staff or other detection devices also learn the information, to carry out corresponding behave.As it can be seen that
Implement this embodiment, the comprehensive of this method can be improved, so that this method effectively be made to be suitable for a variety of different fields
Under scape, and then improve the universality of this method.
The embodiment of the present application second aspect provides a kind of mist detecting device, and the mist detecting device includes:
Acquiring unit, for obtaining Smoke Detection video;
Pretreatment unit obtains processing video for pre-processing to the Smoke Detection video;The processing video
Including multiple processing frames;
First taxon is obtained for carrying out single frames sort operation to each processing frame according to the first intelligent algorithm
To the first classification results;
Second taxon, for including for indicating the first judging result existing for smog in first classification results
When, multiframe sort operation is carried out to the multiple processing frame according to the second intelligent algorithm, obtains the second classification results;
Output unit, for when second classification results include first judging result, output described first to be sentenced
Disconnected result.
During above-mentioned realization, video sequence to be detected can be preferentially obtained using this kind of mist detecting device, and
The pretreatment to video sequence is completed, processing video is obtained;After obtaining processing video, according to the first intelligent algorithm pair
It handles video and carries out single frames classification processing, to complete preliminary smog judgement, according further to the second intelligent algorithm to place
It manages video and carries out multiframe classification processing, to complete accurate smog judgement, so that higher the first judging result of smog of precision is obtained,
And it is somebody's turn to do in last output as a result, so that relevant operator learns testing result.It, can be with as it can be seen that implement this embodiment
The effective detection accuracy for improving Smoke Detection, and the probability to report an error can be substantially reduced by secondary detection, so that
The smog detection method has wider use in practice.
The embodiment of the present application third aspect provides a kind of electronic equipment, including memory and processor, the storage
Device is for storing computer program, and the processor runs the computer program so that the computer equipment is executed according to this
Apply for smog detection method described in any one of embodiment first aspect.
The embodiment of the present application fourth aspect provides a kind of computer readable storage medium, is stored with computer program and refers to
It enables, when the computer program instructions are read and run by a processor, executes any one of the embodiment of the present application first aspect
The smog detection method.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application will make below to required in the embodiment of the present application
Attached drawing is briefly described, it should be understood that the following drawings illustrates only some embodiments of the application, therefore should not be seen
Work is the restriction to range, for those of ordinary skill in the art, without creative efforts, can be with
Other relevant attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of flow diagram of smog detection method provided by the embodiments of the present application;
Fig. 2 is the flow diagram of another smog detection method provided by the embodiments of the present application;
Fig. 3 is a kind of structural schematic diagram of mist detecting device provided by the embodiments of the present application;
Fig. 4 is a kind of schematic diagram of video adjustment provided by the embodiments of the present application;
Fig. 5 is a kind of training process schematic diagram of first artificial intelligence model provided by the embodiments of the present application.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application is described.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.Meanwhile the application's
In description, term " first ", " second " etc. are only used for distinguishing description, are not understood to indicate or imply relative importance.
Embodiment 1
Fig. 1 is please referred to, Fig. 1 provides a kind of flow diagram of smog detection method for the embodiment of the present application.Wherein, should
Smog detection method includes:
S101, Smoke Detection video is obtained.
In the present embodiment, Smoke Detection video can be understood as continuous videos sequence image frame, and in subsequent processing
It is all to be handled based on Smoke Detection video every frame in continuous videos sequence image frame in the process.
In the present embodiment, Smoke Detection video can be real-time detection, be also possible to what user inputted at him, to this
It is not limited in any way in the present embodiment.
In the present embodiment, Smoke Detection video is for detecting whether there are smog, not in the Smoke Detection video
Certainly exist smog.
S102, Smoke Detection video is pre-processed, obtains processing video, processing video includes multiple processing frames.
In the present embodiment, processing video distance be can be understood as in Smoke Detection video, and there are the moving objects of moving object
Volumetric video.Wherein, processing video includes multiple processing frames, which is in processing video.
As an alternative embodiment, being pre-processed to Smoke Detection video, processing video is obtained, video is handled
Include the steps that multiple processing frames include:
Size pretreatment is carried out to Smoke Detection video, obtains size preprocessed video;
Moving object edge processing is carried out to size preprocessed video, obtains processing video, which includes multiple
Handle frame.
Implement this embodiment, pretreated process can be refined as to size pretreatment and moving object edge is located in advance
Manage two steps, it is seen then that pre-processed by above-mentioned two step, pre-processed results can be made more unified, so as to keep away
Exempt from as image disunity and caused by treatment deviation, and then the pretreated process can be made more rigorous, so that pre- place
It manages that obtained pre-processed results are more accurate, and is also possible that the acquisition with processing result is more efficient.And this its
In, the pre-processed results that the pretreated step in moving object edge can be refined further, so that above-mentioned effect
It is more obvious.
In the present embodiment, during obtaining processing video, two pictures can be first established in Smoke Detection video
Frame buffer queue Q1, Q2, wherein illustrative, buffer queue Q1 can be every 2 frame buffer, one frame picture frame, buffer queue Q2
It can be every 20 frame buffer, one frame picture frame;And when the number of image frames that Q1 or Q2 is cached is greater than 8, extracted from corresponding queue
Preceding 8 frame picture frame as the processing video, meanwhile, which is deleted from corresponding buffer queue.
For example, when Smoke Detection video (or above-mentioned size preprocessed video) has 20 frame, Q1 queue
In be stored with 10 frame pictures, 1 frame picture is stored in Q2 queue, among these, obtains the first eight in Q1 queue in 10 frame pictures
Frame forms above-mentioned size preprocessed video, so that the object of edge processing is the size preprocessed video of eight frame, and after the completion
Continuous step.
S103, single frames sort operation is carried out to each processing frame according to the first intelligent algorithm, obtains the first classification knot
Fruit.
In the present embodiment, the intelligent algorithm that the first intelligent algorithm can be constituted for neural network, processing pair
As if each processing frame in above-mentioned processing video, and processing method is the sort operation processing of single frames, and is finally obtaining
Corresponding i.e. the first classification results of single frames processing result, first classification results be used to indicate in the single-frame images of its detection whether
There are smog.
It is single frames disaggregated model using the corresponding model of the first intelligent algorithm, and be using this in the present embodiment
Single frames disaggregated model prediction result step includes: that (the processing video can specifically be interpreted as above-mentioned motion detection to processing video
The matrix of consequence of middle output), which is traversed, the figure in intercepting process video (matrix of consequence)
Picture, and use single frames disaggregated model (can be made of neural network) to derive respectively this eight picture, obtain first point
Class result.In first classification results, if in eight pictures with the presence of an any picture smog image when, by the processing
Video (matrix of consequence) and above-mentioned eight images are exported.If not having any picture in eight pictures, there are smog figures
When picture, then smokeless result is exported.
S104, when the first classification results include for indicating the first judging result existing for smog, it is artificial according to second
Intelligent algorithm carries out multiframe sort operation to multiple processing frames, obtains the second classification results.
In the present embodiment, which is intended to indicate that existing for smog.
In the present embodiment, the intelligent algorithm that the second intelligent algorithm can be constituted for neural network, processing pair
As if all processing frames in above-mentioned processing video, and processing method is the sort operation processing of multiframe, and is finally obtaining
Corresponding entirety classification results i.e. the second classification results, second classification results are used to indicate the processing video (multiframe of its detection
Image) in whether be truly present smog.
In the present embodiment, (multiframe classification can be understood as using the corresponding artificial intelligence model of the second intelligent algorithm
Model) prediction result, the additional processing video of above-mentioned first classification results (matrix of consequence) and above-mentioned eight picture are mentioned
It takes, and the multiframe disaggregated model that this eight picture is made of according to block neural network is derived, export the second classification results,
Second classification results are for judging there is smog or non smoke in processing video.
S105, when the second classification results include the first judging result when, export the first judging result.
In the present embodiment, the classification results that can be understood as multiple image (i.e. processing video) in the second classification results (should
Classification results may include having smog, non smoke even light cigarettes mist, dense smoke mist etc.), and when the second classification results are to have cigarette
When the classification results of mist, it being interpreted as the second classification results includes that (the first judging result is used for above-mentioned first judging result
Indicate smog presence), export the first judging result of smog again at this time.
In the present embodiment, step S101~step S105 method be broadly divided into video acquisition, video pre-filtering (including fortune
Dynamic region detection), several steps of single-frame images classification, multiple image classification and result output.As it can be seen that this method can manage
Solution is is a kind of real-time detection method for video sequence, wherein this method can solve the prior art to video sequence into
The problem that detection accuracy is low when row real-time detection, arithmetic speed is slow.
In the present embodiment, this method and the existing two class Smoke Detection algorithms based on video sequence analysis have bigger difference
It is different, wherein first kind Smoke Detection algorithm is that smoke region is detected and classified using conventional machines learning method, such
Method carries out moving region detection and classification to video sequence by the feature of Manual definition, has the characteristics that arithmetic speed is fast,
But nicety of grading is relatively low in practical applications for the mode of Manual definition's feature that uses of this kind of method, exist a large amount of wrong reports and
The case where misrepresenting deliberately, it is difficult to be applied to complex scene;Second class Smoke Detection algorithm is using neural network method to video sequence
Classify, such method is automatically extracted smoke characteristics using neural network and learnt from training sample, has classification
Feature with high accuracy, but the mode of this kind use neural network speed when carrying out sort operation is slower, it is difficult to reach video
The requirement of real-time detection.As it can be seen that this technique as described in this embodiment is applied to complex scene, reality also may be implemented
When detect, so this method as described in this embodiment precision with higher and universality.
As shown in Figure 1, implementing smog detection method described in Fig. 1, Smoke Detection video can be preferentially obtained, and according to
The Smoke Detection video is pre-processed accordingly, obtains processing video;After this, reason video carries out the first step according to this
Smoke Detection rough sort is carrying out second step Smoke Detection disaggregated classification, so that corresponding second classification results of secondary classification are obtained,
And result information is exported according to second classification results.As it can be seen that implementing this embodiment, can be mentioned by pretreated step
The universality of high this method can also be classified by secondary detection and improve the precision of Smoke Detection, and reduce the rate that reports an error, to make
Obtaining the result information finally exported, there is very high confidence level to mention so that this method can be adapted for more scenes
The high universality of Smoke Detection.
Embodiment 2
Fig. 2 is please referred to, Fig. 2 is the flow diagram of another smog detection method provided by the embodiments of the present application.Fig. 2 institute
The flow diagram of the smog detection method of description is that the flow diagram of the smog detection method according to described in Fig. 1 is changed
Into what is obtained.Wherein, which includes:
S201, Smoke Detection video is obtained.
In the present embodiment, other embodiments can be referred to for identical explanation, to no longer more in this present embodiment
Add and repeats.
In the present embodiment, Smoke Detection video is the detection video for judging whether there is smog.
In the present embodiment, " video " once for indicating processed object, among these, although in concrete processing procedure
Process object should be the picture frame of each video, but after processing is complete, picture frame can still form video and (only should
The frame per second of video is lower than normal frame rate), therefore referred to herein as " video ".
In the present embodiment, corresponding Smoke Detection video is picture frame, and it is processing frame, motion detection that it is corresponding, which to handle video,
Corresponding video is video frame.
S202, size adjusting is carried out to each picture frame in Smoke Detection video according to pre-set dimension, after being adjusted
Adjustment video.
In the present embodiment, pre-set dimension can be understood as resolution ratio it can be appreciated that physical size, to this present embodiment
In be not limited in any way.
For example, right first after inputting Smoke Detection video (can be understood as continuous videos sequence image frame)
The filling of ater background edge is carried out to picture frame each in the Smoke Detection video, so that each picture frame length-width ratio is 16:
9, so that the length-width ratio of Smoke Detection video is 16:9;Then image resolution ratio is zoomed into 1280*720 again, to make
It obtains Smoke Detection video and is integrally adjusted to adjustment video.
Referring to Fig. 4, as shown in figure 4, Smoke Detection video is adjusted be adjusted the schematic diagram of video can be straight
See the acquisition process that adjustment video is found out on ground.
S203, the extraction for carrying out video frame to adjustment video according to default detection mode, and with the video frame of extraction
For according to acquisition motion detection video.
In the present embodiment, moving region detection is the detection to whether there is moving object in adjustment video.Wherein, the fortune
Dynamic detection video is for detecting whether being truly present the detection video of moving object.
In the present embodiment, during obtaining adjustment video, it is slow two picture frames first can be established in adjustment video
Queue Q1, Q2 are deposited, illustratively, buffer queue Q1 can be every 2 frame buffer, one frame picture frame, and buffer queue Q2 can be every 20
One frame picture frame of frame buffer;And when the number of image frames that Q1 or Q2 is cached is greater than 8, preceding 8 frame image is extracted from corresponding queue
Frame is used as processing video, meanwhile, which is deleted from corresponding buffer queue.
For example, when adjusting video has 20 frame, 10 frame pictures is stored in Q1 queue, are stored with 1 in Q2 queue
Frame picture obtains the first eight frame in Q1 queue in 10 frame pictures, above-mentioned size preprocessed video is formed, so that side among these
The object of edge processing is the size preprocessed video of eight frame, and completes subsequent step.
In the present embodiment, the step of moving region is detected further include: 8 frame images of said extracted are carried out in sequence
F0, f1, f2, f3, f4, f5, f6, f7 are named, and extracts first frame image f0, the 4th frame image f3 and the 8th frame image f7, it
Afterwards, which is quantized, and seeks the matrix of differences d2 of the matrix of differences d1 and f3 and f7 of f0 and f3, as follows:
D1=| f3-f0 |;
D2=| f7-f3 |.
S204, judge with the presence or absence of moving object in motion detection video, if so, thening follow the steps S205~S208;If
It is no, then terminate this process.
For example binaryzation, then to d1 and d2 is carried out, when numerical value is greater than certain threshold value in matrix, when being greater than 30,
Think that it, for foreground moving object, is assigned a value of 1;When numerical value is less than 30, it is believed that it is background, is assigned a value of 0.Wherein, work as assignment
It when being 1, indicates that there are moving objects, when being assigned a value of 0, indicates that moving object is not present.
S205, the contour edge of moving object is extracted in motion detection video, obtains processing video, processing view
Frequency includes multiple processing frames.
In the present embodiment, processing frame is picture frame, is to handle the content frame for including in video.
For example, which can be used canny operator and carries out foreground moving object wheel to above-mentioned binarization result
Wide edge extracting, and the external matrix that profile calculates profile is traversed, and export 8 frame images (above-mentioned multiple processing frames) and corresponding
All external matrixes.Wherein, eight above-mentioned frame images and corresponding all external matrixes can be understood as processing video.
S206, single frames sort operation is carried out to each processing frame according to the first intelligent algorithm, obtains the first classification knot
Fruit.
As an alternative embodiment, corresponding first artificial intelligence model of the first intelligent algorithm includes five
Convolutional layer, the first full articulamentum and the first classification layer;Wherein, each processing frame is carried out according to the first intelligent algorithm single
Frame classification operation, the step of obtaining the first classification results include:
Convolution algorithm is carried out to each processing frame according to five convolutional layers, obtains the first characteristics of image group;
Data processing is carried out to the first characteristics of image group according to the first full articulamentum, obtains first eigenvector group;
Classified according to the first classification layer to first eigenvector group, obtains the first classification results.
Implement this embodiment, can refine and processing video is handled to obtain the specific mistake of the first classification results
Journey, and can also determine the specific structure of the first artificial intelligence model, so that processing video can obtain with this configuration
To accurate first classification results, and then improve the classification accuracy of the first classification results;Meanwhile the process is to carry out single frames
Detection, that is to say, that its detection range is bigger in this process, and effect is more preferable, and accuracy is also higher.
As further alternative embodiment, the first classification layer includes first-loss function;Wherein, according to first point
After the step of class layer classifies to first eigenvector group, obtains the first classification results, method further include:
The first classification results are calculated according to first-loss function, obtain the first difference value;
The sorting parameter that the first classification layer includes is updated according to the first difference value and preset stochastic gradient descent algorithm.
Implement this embodiment, the step similar with training process can be applied with it is actually detected in, realize real
The result negative-feedback of border detection, so that the model is able to carry out autonomous iteration, to independently optimize its function.
In the present embodiment, which can be understood as carrying out rough sort to moving region using single frames smog sorting algorithm.
Wherein, the establishment process of corresponding first artificial intelligence model of the first intelligent algorithm includes that data are manually demarcated, single frames divides
Class model training, the prediction of single frames disaggregated model.Detailed process is as follows: single frames sample database TS is established, to defeated in motion detection
F0 ... f7 out intercepts the image within the scope of matrix, obtains corresponding multiple images block, and TS is added in these image blocks;
When TS scale reaches certain amount (such as 200000), using the mode manually determined, smog is made whether to image block
Determine, when being that there are the labels for when smog, setting corresponding image block in above-mentioned image block as T, otherwise sets label as F;
It initializes neural network single frames disaggregated model (i.e. the first artificial intelligence model).The mode input leads to for the 3 of 640*480 resolution ratio
Road characteristics of image figure;By the convolutional layer C0 ... C4 and full connection F5 of five different scales, feature is extracted to characteristic pattern and is obtained
Feature vector;Classification results are finally obtained to feature vector recurrence using Softmax regression function.
In the present embodiment, the model specific structure and function of the first artificial intelligence model be can be such that
C0 convolutional layer: convolution algorithm is carried out using 7*7*3 convolution kernel and 3 channel characteristics figures, to operation result application ReLU
The activation of line rectification function, ReLU function formula are as follows:
F (x)=max (0, x);
It is down-sampled using pondization operation progress to activation result, make to export characteristics of image channel 64;
C1 convolutional layer: convolutional calculation is carried out to C0 output characteristics of image using 3*3*3 convolution kernel, keeps output characteristics of image logical
Road is 64;
C2 convolutional layer: convolutional calculation is carried out to C1 output characteristics of image using 3*3*3 convolution kernel, keeps output characteristics of image logical
Road is 128;
C3 convolutional layer: convolutional calculation is carried out to C3 output characteristics of image using 3*3*3 convolution kernel, keeps output characteristics of image logical
Road is 256;
C4 convolutional layer: convolutional calculation is carried out to C4 output characteristics of image using 3*3*3 convolution kernel, keeps output characteristics of image logical
Road is 512;
Full articulamentum F5: using 3*3*1*512 full attended operation by the C4 result exported be converted into 512 features to
Measure Feat;
Classification results calculate: feature vector calculated using softmax function, the calculating function of softmax is as follows:
Wherein z is Feat vector (being suitble to vector for indicating), and K is the dimension of Feat vector, and 512 bit vectors are calculated
The each element ∈ (0,1) of σ, σ, and all elements and be 1.
In the present embodiment, the specific steps of training single frames disaggregated model (training the first artificial intelligence model) can be as
Under:
By 200000 pictures scaling in TS and fixed ratio is filled into 640*480 resolution ratio, and is divided into two groups, wherein
The ratio of training group and validation group is 9:1;
The parameter of disaggregated model is initialized by the way of random initializtion;
Using stochastic gradient descent (SGD) method training pattern, learning rate in SGD method can be set based on experience value and is joined
Number be 0.01, momentum parameter 0.9, every 40,000 iteration adjustment of learning rate be its 1/10.Using softmax loss function:
Wherein LiIt is in predicted vector (can be understood as the obtained result picture of classification, i.e., corresponding first classification results) the
The difference of i element and label vector (having manually added the picture of label in advance) i-th of element, λ W are constant value;If point
Class is correct, then loss function value is constant, and otherwise loss function value changes;
Picture every in training set is calculated using neural network single frames disaggregated model, exports classification prediction result,
Classification entrance loss in classification and label is calculated into function and obtains difference value, difference value is returned and is updated according to SGD method and divided
Parameter in class model.Continuous iteration process simultaneously calculates the classification results on verifying collection, the nicety of grading on verifying collection
Reach 99%, deconditioning process exports parameter matrix M1 at this time.
As it can be seen that the first above-mentioned artificial intelligence model can execute corresponding step during foundation, and complete
After the foundation of first artificial intelligence model, effectively further execute single frames sort operation the step of.
Referring to Fig. 5, as shown in figure 5, the training process of single frames disaggregated model (i.e. the first artificial intelligence model was trained
Journey) it can intuitively show.Wherein, content shown in fig. 5 is the training process of the first artificial intelligence model, wherein first
The artificial intelligence model correspondence single frames disaggregated model.Meanwhile second artificial intelligence model training process it is similar therewith.
It include that there are also other parameters, sorting parameters described herein for loss function in the present embodiment, in the first classification layer
It is two concepts with first-loss function.That is, what is updated here is the parameter in the first classification layer, it is not loss letter
Number.In addition, if if parameter in first-loss function, it should be expressed as the bottom loss function in the first classification layer included
Parameter, and it is said herein be the first classification layer parameter, here it can be seen that the rank of its parameter is not identical, therefore cannot
Enough mention in the same breath.Meanwhile explanation is used to classify sorting parameter at last, and loss function has differences with this, therefore two
Person does not conflict.Based on this, subsequent second classification layer is explained identical.
S207, when the first classification results include for indicating the first judging result existing for smog, it is artificial according to second
Intelligent algorithm carries out multiframe sort operation to multiple processing frames, obtains the second classification results.
As an alternative embodiment, corresponding second artificial intelligence model of the second intelligent algorithm includes three
Convolutional layer, the second full articulamentum and the second classification layer;Wherein, multiple processing frames are carried out according to the second intelligent algorithm more
Frame classification operation, the step of obtaining the second classification results include:
According to three convolutional layers convolution algorithm is carried out to multiple processing frames respectively, obtains the second characteristics of image group;
Data processing is carried out to the second characteristics of image group according to the second full articulamentum, obtains second feature Vector Groups;
Classified according to the second classification layer to second feature Vector Groups, obtains the second classification results.
Implement this embodiment, can refine and processing video is handled to obtain the specific mistake of the second classification results
Journey, and can also determine the specific structure of the second artificial intelligence model, so that processing video can obtain with this configuration
To accurate second classification results, and then improve the classification accuracy of the second classification results;Meanwhile the process is to carry out multiframe
Detection, that is to say, that its detection accuracy is higher in this process, and effect is more preferable, can effectively carry out smog high-precision
Detection;Also, the process can be used as secondary detection presence, so that there are smog in the testing result obtained after one-time detection
When realize high-precision judgement, so that the raising for the accuracy that Smoke Detection integrally judges is realized, so that this kind of height
The method of precision can be useful under various environment, improve its universality.
As a kind of further alternative embodiment, the second classification layer includes the second loss function;Wherein, according to
After the step of two classification layers classify to second feature Vector Groups, obtain the second classification results, method further include:
The second classification results are calculated according to the second loss function, obtain the second difference value;
The sorting parameter that the second classification layer includes is updated according to the second difference value and preset stochastic gradient descent algorithm.
Implement this embodiment, the step similar with training process can be applied with it is actually detected in, realize real
The result negative-feedback of border detection, so that the model is able to carry out autonomous iteration, to independently optimize its function.
In the present embodiment, above-mentioned step is using multiframe smog sorting algorithm (i.e. the second artificial intelligence model) to single frames
The step of result of smog classification is finely divided class, wherein the establishment step of the second artificial intelligence model includes that data are manually marked
Fixed, single frames disaggregated model training, the prediction of single frames disaggregated model.Specific step is as follows: multiframe sample database TM is established, to rough segmentation
The matrix exported in class intercepts the multiple images block of f0 ... f7 within the scope of matrix to each matrix, and by these image blocks
Group is combined into a sequence blocks, then TM is added in sequence blocks;When TM scale reaches certain amount (such as 200000 groups), using artificial
The mode of judgement is made whether the judgement of smog to sequence blocks.When there are when smog, set the image block sequence for image block sequence
The label of column is T, otherwise sets label as F;It initializes neural network multiframe disaggregated model (i.e. the second artificial intelligence model).It should
Mode input is 3 channel image characteristic patterns of continuous 8 frame 112*112 resolution ratio;And the convolutional layer C0 of 3 different scales is used,
C1, C2 carry out feature extraction to characteristic pattern;Characteristic pattern is mapped as feature vector using full articulamentum F3;Finally use
Softmax regression function obtains classification results to feature vector recurrence.
In the present embodiment, the specific structure of the second artificial intelligence model and function can be such that
C0 convolutional layer: feature extraction is carried out to the input of 8*112*112*3 using 3 dimension convolution operations, convolution kernel size is
3*3*3, output characteristics of image channel are 64;
C1 convolutional layer: same to C0, output characteristics of image channel are 128;
C2 convolutional layer: same to C1, output characteristics of image channel are 256;
The full articulamentum of F3: the C2 result exported is converted to using the full attended operation of 3*,3*1,*8*,256 256 features
Vector;
Classification results calculate: calculating using softmax function feature vector, obtain classification results.
In the present embodiment, the specific steps of training multiframe disaggregated model (train the second artificial intelligence model) with it is above-mentioned
The first artificial intelligence model of training is similar, wherein needs the resolution adjustment of data set picture to be 112*112, and same
When adjustment input be sequence of pictures.
S208, when the second classification results include the first judging result when, export the first judging result;When the second classification results
When not including the first judging result, the second judging result for indicating non smoke is exported.
In the present embodiment, the first judging result and the second judging result are result arranged side by side.
As it can be seen that implementing smog detection method described in Fig. 2, Smoke Detection video can be preferentially obtained, and according to the cigarette
Mist detection video is pre-processed accordingly, obtains processing video;After this, reason video carries out first step smog according to this
Rough sort is detected, second step Smoke Detection disaggregated classification is being carried out, to obtain corresponding second classification results of secondary classification, and root
Result information is exported according to second classification results.As it can be seen that implementing this embodiment, can be improved by pretreated step should
The universality of method can also be classified by secondary detection and improve the precision of Smoke Detection, and reduce the rate that reports an error, so that most
There is the result information exported afterwards very high confidence level to improve so that this method can be adapted for more scenes
The universality of Smoke Detection.
Embodiment 3
Fig. 3 is please referred to, Fig. 3 is a kind of structural schematic diagram of mist detecting device provided by the embodiments of the present application.Wherein, should
Mist detecting device includes:
Acquiring unit 310, for obtaining Smoke Detection video;
Pretreatment unit 320 obtains processing video for pre-processing to Smoke Detection video;Handling video includes
Multiple processing frames;
First taxon 330, for carrying out single frames sort operation to each processing frame according to the first intelligent algorithm,
Obtain the first classification results;
Second taxon 340, for including for indicating the first judging result existing for smog in the first classification results
When, multiframe sort operation is carried out to multiple processing frames according to the second intelligent algorithm, obtains the second classification results;
Output unit 350, for exporting the first judging result when the second classification results include the first judging result.
In the present embodiment, mist detecting device can execute specific step described in above-described embodiment 1 or embodiment 2
Suddenly, to no longer adding to repeat in this present embodiment.
As an alternative embodiment, pretreatment unit 320 may include size adjusting subelement, motion detection
Unit, judgment sub-unit and extraction subelement, wherein
Size adjusting subelement, for carrying out size tune to each picture frame in Smoke Detection video according to pre-set dimension
It is whole, the adjustment video after being adjusted;
Motion detection subelement, for carrying out the extraction of video frame to adjustment video according to default detection mode, and to take out
The video frame taken is according to acquisition motion detection video;
Judgment sub-unit, for judging in motion detection video with the presence or absence of moving object;
Subelement is extracted, is when being, to moving object in motion detection video for the judging result in judgment sub-unit
The contour edge of body extracts, and obtains processing video.
Implement this embodiment, can refine with 320 specific structure of processing unit, to realize corresponding function.
As an alternative embodiment, corresponding first artificial intelligence model of the first intelligent algorithm includes five
Convolutional layer, the first full articulamentum and the first classification layer;Wherein, the first taxon 330 includes the first convolution subelement, first
Full connection unit and the first taxon, wherein
First convolution subelement obtains the first figure for carrying out convolution algorithm to each processing frame according to five convolutional layers
As feature group;
First full connection unit is obtained for carrying out data processing to the first characteristics of image group according to the first full articulamentum
First eigenvector group;
First taxon obtains the first classification for classifying according to the first classification layer to first eigenvector group
As a result.
Implement this embodiment, the first intelligent algorithm can be refined, so that the first taxon 330 can be with
Classification processing is carried out according to first intelligent algorithm, obtains the first classification results.
As an alternative embodiment, the first classification layer includes first-loss function;Wherein, the first taxon
330 can also include that the first computation subunit and first update subelement, wherein
It is poor to obtain first for calculating according to first-loss function the first classification results for first computation subunit
Different value;
First updates subelement, for updating the first classification according to the first difference value and preset stochastic gradient descent algorithm
The sorting parameter that layer includes.
Implement this embodiment, self feed back may be implemented, to improve the classification capacity of the first classification layer, and then can be with
Improve nicety of grading.
As an alternative embodiment, corresponding second artificial intelligence model of the second intelligent algorithm includes three
Convolutional layer, the second full articulamentum and the second classification layer;Wherein, the second taxon 340 may include the second convolution subelement,
Second full connection unit and the second taxon, wherein
Second convolution subelement obtains for carrying out convolution algorithm respectively to multiple processing frames according to three convolutional layers
Two characteristics of image groups;
Second full connection unit is obtained for carrying out data processing to the second characteristics of image group according to the second full articulamentum
Second feature Vector Groups;
Second taxon obtains the second classification for classifying according to the second classification layer to second feature Vector Groups
As a result.
As an alternative embodiment, the second classification layer includes the second loss function;Wherein, the second taxon
340 can also include that the second computation subunit and second update subelement, wherein
It is poor to obtain second for calculating according to the second loss function the second classification results for second computation subunit
Different value;
Second updates subelement, for updating the second classification according to the second difference value and preset stochastic gradient descent algorithm
The sorting parameter that layer includes.
As an alternative embodiment, output unit 350 is also used to when the second classification results not include the first judgement
When as a result, the second judging result for indicating non smoke is exported.
As it can be seen that mist detecting device described in implementing Fig. 3, can preferentially obtain video sequence to be detected, and complete
Pretreatment to video sequence obtains processing video;After obtaining processing video, according to the first intelligent algorithm to processing
Video carries out single frames classification processing, to complete preliminary smog judgement, regards according further to the second intelligent algorithm to processing
Frequency carries out multiframe classification processing, to complete accurate smog judgement, so that higher the first judging result of smog of precision is obtained, and
Finally output should be as a result, so that relevant operator learns testing result.As it can be seen that implement this embodiment, it can be effective
Raising Smoke Detection detection accuracy, and the probability to report an error can be substantially reduced by secondary detection, so that the cigarette
Mist detection method has wider use in practice.
The embodiment of the present application also provides a kind of electronic equipment, which includes memory and processor, institute
Memory is stated for storing computer program, the processor runs the computer program so that the computer equipment executes
The smog detection method of any one of 1~embodiment of embodiment 2 description.
The embodiment of the present application also provides a kind of computer readable storage mediums, are stored in above-mentioned computer equipment
Used computer program (or smog detection method of any one of execution 1~embodiment of embodiment 2 description).
In several embodiments provided herein, the same or similar noun limited, illustrated, step limits
Fixed, step addition, operation limit and identical explanation all can be used in operation addition, meanwhile, because of Smoke Detection dress
Set corresponding with smog detection method, therefore corresponding illustrate can also correspond to reference explanation, in this present embodiment not
It is repeated again.
In several embodiments provided herein, it should be understood that disclosed device and method can also pass through
Other modes are realized.The apparatus embodiments described above are merely exemplary, for example, flow chart and block diagram in attached drawing
Show the device of multiple embodiments according to the application, the architectural framework in the cards of method and computer program product,
Function and operation.In this regard, each box in flowchart or block diagram can represent the one of a module, section or code
Part, a part of the module, section or code, which includes that one or more is for implementing the specified logical function, to be held
Row instruction.It should also be noted that function marked in the box can also be to be different from some implementations as replacement
The sequence marked in attached drawing occurs.For example, two continuous boxes can actually be basically executed in parallel, they are sometimes
It can execute in the opposite order, this depends on the function involved.It is also noted that every in block diagram and or flow chart
The combination of box in a box and block diagram and or flow chart can use the dedicated base for executing defined function or movement
It realizes, or can realize using a combination of dedicated hardware and computer instructions in the system of hardware.
In addition, each functional module in each embodiment of the application can integrate one independent portion of formation together
Point, it is also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It, can be with if the function is realized and when sold or used as an independent product in the form of software function module
It is stored in a computer readable storage medium.Based on this understanding, the technical solution of the application is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) execute each embodiment the method for the application all or part of the steps.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
The above description is only an example of the present application, the protection scope being not intended to limit this application, for ability
For the technical staff in domain, various changes and changes are possible in this application.Within the spirit and principles of this application, made
Any modification, equivalent substitution, improvement and etc. should be included within the scope of protection of this application.It should also be noted that similar label and
Letter indicates similar terms in following attached drawing, therefore, once it is defined in a certain Xiang Yi attached drawing, then in subsequent attached drawing
In do not need that it is further defined and explained.
The above, the only specific embodiment of the application, but the protection scope of the application is not limited thereto, it is any
Those familiar with the art within the technical scope of the present application, can easily think of the change or the replacement, and should all contain
Lid is within the scope of protection of this application.Therefore, the protection scope of the application shall be subject to the protection scope of the claim.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Claims (10)
1. a kind of smog detection method characterized by comprising
Obtain Smoke Detection video;
The Smoke Detection video is pre-processed, obtains processing video, the processing video includes multiple processing frames;
Single frames sort operation is carried out to each processing frame according to the first intelligent algorithm, obtains the first classification results;
When first classification results include for indicating the first judging result existing for smog, calculated according to the second artificial intelligence
Method carries out multiframe sort operation to the multiple processing frame, obtains the second classification results;
When second classification results include first judging result, first judging result is exported.
2. smog detection method according to claim 1, which is characterized in that described to be carried out in advance to the Smoke Detection video
Processing, obtain processing video the step of include:
Size adjusting is carried out to each picture frame in the Smoke Detection video according to pre-set dimension, the adjustment after being adjusted
Video;
The extraction of video frame is carried out to the adjustment video according to default detection mode, and using the video frame of extraction as foundation
Obtain motion detection video;
Judge in the motion detection video with the presence or absence of moving object;
If there are when the moving object in the motion detection video, to the moving object in the motion detection video
Contour edge extract, obtain processing video.
3. smog detection method according to claim 1, which is characterized in that first intelligent algorithm corresponding
One artificial intelligence model includes five convolutional layers, the first full articulamentum and the first classification layer;Wherein, described artificial according to first
Intelligent algorithm carries out single frames sort operation to each processing frame, and the step of obtaining the first classification results includes:
Convolution algorithm is carried out to each processing frame according to five convolutional layers, obtains the first characteristics of image group;
Data processing is carried out to the first image feature group according to the described first full articulamentum, obtains first eigenvector group;
Classified according to the first classification layer to the first eigenvector group, obtains first classification results.
4. smog detection method according to claim 3, which is characterized in that the first classification layer includes first-loss letter
Number;Wherein, classified according to the first classification layer to the first eigenvector group described, obtain the first classification results
The step of after, the method also includes:
First classification results are calculated according to the first-loss function, obtain the first difference value;
The classification ginseng that the first classification layer includes is updated according to first difference value and preset stochastic gradient descent algorithm
Number.
5. smog detection method according to claim 1, which is characterized in that second intelligent algorithm corresponding
Two artificial intelligence models include three convolutional layers, the second full articulamentum and the second classification layer;Wherein, described artificial according to second
Intelligent algorithm carries out multiframe sort operation to the multiple processing frame, and the step of obtaining the second classification results includes:
According to three convolutional layers convolution algorithm is carried out to the multiple processing frame respectively, obtains the second characteristics of image group;
Data processing is carried out to the second characteristics of image group according to the described second full articulamentum, obtains second feature Vector Groups;
Classified according to the second classification layer to the second feature Vector Groups, obtains second classification results.
6. smog detection method according to claim 5, which is characterized in that the second classification layer includes the second loss letter
Number;Wherein, classified according to the second classification layer to the second feature Vector Groups described, obtain the second classification results
The step of after, the method also includes:
Second classification results are calculated according to second loss function, obtain the second difference value;
The classification ginseng that the second classification layer includes is updated according to second difference value and preset stochastic gradient descent algorithm
Number.
7. smog detection method according to claim 1, which is characterized in that the method also includes:
When second classification results do not include first judging result, the second judgement knot for indicating non smoke is exported
Fruit.
8. a kind of mist detecting device, which is characterized in that the mist detecting device includes:
Acquiring unit, for obtaining Smoke Detection video;
Pretreatment unit obtains processing video for pre-processing to the Smoke Detection video;The processing video includes
Multiple processing frames;
First taxon, for, to each processing frame progress single frames sort operation, obtaining the according to the first intelligent algorithm
One classification results;
Second taxon, for when first classification results include for indicating the first judging result existing for smog,
Multiframe sort operation is carried out to the multiple processing frame according to the second intelligent algorithm, obtains the second classification results;
Output unit, for when second classification results include first judging result, exporting the first judgement knot
Fruit.
9. a kind of electronic equipment, which is characterized in that the electronic equipment includes memory and processor, and the memory is used for
Computer program is stored, the processor runs the computer program so that the computer equipment is executed according to claim
Smog detection method described in any one of 1 to 7.
10. a kind of readable storage medium storing program for executing, which is characterized in that computer program instructions are stored in the read/write memory medium,
When the computer program instructions are read and run by a processor, perform claim requires 1 to 7 described in any item Smoke Detections
Method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910236445.2A CN110070106A (en) | 2019-03-26 | 2019-03-26 | Smog detection method, device and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910236445.2A CN110070106A (en) | 2019-03-26 | 2019-03-26 | Smog detection method, device and electronic equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110070106A true CN110070106A (en) | 2019-07-30 |
Family
ID=67366698
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910236445.2A Pending CN110070106A (en) | 2019-03-26 | 2019-03-26 | Smog detection method, device and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110070106A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112052744A (en) * | 2020-08-12 | 2020-12-08 | 成都佳华物链云科技有限公司 | Environment detection model training method, environment detection method and device |
CN112216351A (en) * | 2020-09-23 | 2021-01-12 | 成都佳华物链云科技有限公司 | Desulfurization optimization method and device, electronic equipment and storage medium |
CN112437279A (en) * | 2020-11-23 | 2021-03-02 | 方战领 | Video analysis method and device |
CN112989934A (en) * | 2021-02-05 | 2021-06-18 | 方战领 | Video analysis method, device and system |
DE102021116957A1 (en) | 2021-07-01 | 2023-01-05 | IQ Technologies for Earth and Space GmbH | Method and device for the automated early detection of wildfires by means of optical detection of smoke events |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103983574A (en) * | 2014-06-03 | 2014-08-13 | 上海安维尔信息科技有限公司 | Smoke detection method |
CN106650600A (en) * | 2016-10-17 | 2017-05-10 | 东南大学 | Forest smoke and fire detection method based on video image analysis |
CN107633212A (en) * | 2017-08-30 | 2018-01-26 | 清华大学苏州汽车研究院(吴江) | A kind of firework detecting method and device based on video image |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
CN109409256A (en) * | 2018-10-10 | 2019-03-01 | 东南大学 | A kind of forest rocket detection method based on 3D convolutional neural networks |
-
2019
- 2019-03-26 CN CN201910236445.2A patent/CN110070106A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103983574A (en) * | 2014-06-03 | 2014-08-13 | 上海安维尔信息科技有限公司 | Smoke detection method |
CN106650600A (en) * | 2016-10-17 | 2017-05-10 | 东南大学 | Forest smoke and fire detection method based on video image analysis |
CN107633212A (en) * | 2017-08-30 | 2018-01-26 | 清华大学苏州汽车研究院(吴江) | A kind of firework detecting method and device based on video image |
CN107749067A (en) * | 2017-09-13 | 2018-03-02 | 华侨大学 | Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks |
CN109409256A (en) * | 2018-10-10 | 2019-03-01 | 东南大学 | A kind of forest rocket detection method based on 3D convolutional neural networks |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112052744A (en) * | 2020-08-12 | 2020-12-08 | 成都佳华物链云科技有限公司 | Environment detection model training method, environment detection method and device |
CN112052744B (en) * | 2020-08-12 | 2024-02-09 | 成都佳华物链云科技有限公司 | Environment detection model training method, environment detection method and environment detection device |
CN112216351A (en) * | 2020-09-23 | 2021-01-12 | 成都佳华物链云科技有限公司 | Desulfurization optimization method and device, electronic equipment and storage medium |
CN112216351B (en) * | 2020-09-23 | 2024-02-13 | 成都佳华物链云科技有限公司 | Desulfurization optimization method and device, electronic equipment and storage medium |
CN112437279A (en) * | 2020-11-23 | 2021-03-02 | 方战领 | Video analysis method and device |
CN112989934A (en) * | 2021-02-05 | 2021-06-18 | 方战领 | Video analysis method, device and system |
CN112989934B (en) * | 2021-02-05 | 2024-05-24 | 方战领 | Video analysis method, device and system |
DE102021116957A1 (en) | 2021-07-01 | 2023-01-05 | IQ Technologies for Earth and Space GmbH | Method and device for the automated early detection of wildfires by means of optical detection of smoke events |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948425B (en) | Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching | |
CN110070106A (en) | Smog detection method, device and electronic equipment | |
CN108537136B (en) | Pedestrian re-identification method based on attitude normalization image generation | |
CN107609525B (en) | Remote sensing image target detection method for constructing convolutional neural network based on pruning strategy | |
CN109670528B (en) | Data expansion method facing pedestrian re-identification task and based on paired sample random occlusion strategy | |
CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
CN111126202A (en) | Optical remote sensing image target detection method based on void feature pyramid network | |
CN105809672B (en) | A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring | |
CN110020592A (en) | Object detection model training method, device, computer equipment and storage medium | |
CN106228539B (en) | A variety of geometric primitive automatic identifying methods in a kind of three-dimensional point cloud | |
CN109284779A (en) | Object detection method based on deep full convolution network | |
CN110009614A (en) | Method and apparatus for output information | |
CN105243356B (en) | A kind of method and device that establishing pedestrian detection model and pedestrian detection method | |
CN110399895A (en) | The method and apparatus of image recognition | |
CN114332578A (en) | Image anomaly detection model training method, image anomaly detection method and device | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN109871821A (en) | The pedestrian of adaptive network recognition methods, device, equipment and storage medium again | |
CN109886356A (en) | A kind of target tracking method based on three branch's neural networks | |
CN106682681A (en) | Recognition algorithm automatic improvement method based on relevance feedback | |
CN110472572A (en) | The quick identification and classification method of naval target under a kind of complex environment | |
CN109284760A (en) | A kind of furniture detection method and device based on depth convolutional neural networks | |
CN109492596A (en) | A kind of pedestrian detection method and system based on K-means cluster and region recommendation network | |
CN110135237B (en) | Gesture recognition method | |
CN110008853A (en) | Pedestrian detection network and model training method, detection method, medium, equipment | |
CN108133235A (en) | A kind of pedestrian detection method based on neural network Analysis On Multi-scale Features figure |
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 |
Application publication date: 20190730 |
|
RJ01 | Rejection of invention patent application after publication |