CN109858516A - A kind of fire and smog prediction technique, system and medium based on transfer learning - Google Patents
A kind of fire and smog prediction technique, system and medium based on transfer learning Download PDFInfo
- Publication number
- CN109858516A CN109858516A CN201811583508.3A CN201811583508A CN109858516A CN 109858516 A CN109858516 A CN 109858516A CN 201811583508 A CN201811583508 A CN 201811583508A CN 109858516 A CN109858516 A CN 109858516A
- Authority
- CN
- China
- Prior art keywords
- fire
- smog
- training
- scene image
- prediction
- 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
Landscapes
- Image Analysis (AREA)
- Fire-Detection Mechanisms (AREA)
Abstract
The present invention relates to a kind of fire based on transfer learning and smog prediction technique, system and medium, method includes obtaining multiple scene image samples comprising fire and smog, and make data set according to the scene image sample;Based on transfer learning method, the pre-training model obtained in advance is trained using the data set, constructs trained target prediction model;Scene image to be measured is predicted according to the target prediction model, obtains the prediction classification of the scene image to be measured.Fire and smog prediction technique of the invention is not necessarily to a large amount of training sample, it can obtain preferable learning effect, and the training time greatly shortens, training cost substantially reduces, realize the classifying quality of the higher precision to the prediction classification of fire and smog, target prediction model is improved to the precision of prediction and forecasting efficiency of the prediction classification of fire and smog, to easily and timely find fire and smog alert, avoids the safety of life and property for threatening people.
Description
Technical field
The present invention relates to fire and smog electric powder prediction more particularly to a kind of fire and smog based on transfer learning
Prediction technique, system and medium.
Background technique
At present, fire and smog have been the significant threats in life at home and industrial production.It is especially raw in industry
In production, fire and dangerous smog are timely and accurately found, grasp fire behavior and smog alert is of crucial importance.So industry is raw
Production has urgently needed solution and can precisely, reliably, rapidly detect to fire and smog.
Mainly production environment is monitored using monitoring camera in industrial production.When fire occurs, prison can be passed through
Control camera shooting shoots environment, and the resulting image of shooting is analyzed and classified by artificial neural network.Depth
Habit is the research using artificial Neural Network Simulation human brain, it by artificial neural network to input data carry out feature extraction,
Representative learning, and can be realized the classification of data, the functions such as abstract.
In conventional methods where it is generally necessary to using mass data repetitive exercise artificial neural network, there is training sample demand
It is more, the training time is long, training is at high cost, the problems such as training effect is difficult to ensure.
Summary of the invention
The technical problem to be solved by the present invention is to solve the above shortcomings of the prior art and to provide one kind to be based on transfer learning
Fire and smog prediction technique, system and medium.
The technical scheme to solve the above technical problems is that
A kind of fire and smog prediction technique based on transfer learning, comprising the following steps:
Step 1: obtaining multiple scene image samples comprising fire and smog, and made according to the scene image sample
Data set;
Step 2: it is based on transfer learning method, the pre-training model obtained in advance is trained using the data set,
Construct trained target prediction model;
Step 3: scene image to be measured being predicted according to the target prediction model, obtains the scene image to be measured
Prediction classification.
The beneficial effects of the present invention are: the convolutional neural networks mould obtained in advance can be inherited due to by transfer learning
Extraction and learning ability of the type to feature, therefore the application is based on transfer learning method, the number made by scene image sample
The pre-training model obtained in advance is trained according to collection, fire that pre-training model concentrates data and smog can be inherited
The extraction and learning ability of feature, can adapt to the demand of small sample training, be not necessarily to a large amount of training sample, it will be able to obtain compared with
Good learning effect, and the training time greatly shortens, training cost substantially reduces, the more conducively feature extraction to fire and smog
And representative learning, the classifying quality of the higher precision to the prediction classification of fire and smog is realized, to improve target prediction mould
Type is to the prediction accuracy and forecasting efficiency of the prediction classification of fire and smog, to easily and timely find that fire and smog are alert
Feelings avoid the safety of life and property for threatening people;
Wherein, pre-training model can be the convolutional neural networks model about fire and smog trained before oneself
Achievement is also possible to other people the convolutional neural networks Model Results about fire and smog.
Based on the above technical solution, the present invention can also be improved as follows:
Further, in the step 1, the specific steps for making the data set include:
Step 11: the scene image sample being pre-processed, target scene image is obtained;
Step 12: extracting the fire and smoke data of the target scene image, and according to the fire and smoke data
Determine the corresponding scene type;
Step 13: the data set is made according to the fire and smoke data and the corresponding scene type.
The beneficial effect of above-mentioned further scheme is: there are many pretreatment modes, passes through the pre- place to scene image sample
Scene image sample, can be unified into same substandard image that target prediction model can be supported, such as at normalization by reason
Reason, can also improve the overall contrast of scene image sample, such as histogram equalization processing, can also be by scene image sample
Fire and smoke data in this, which mark out, to be come, convenient for the fire and smoke data in subsequent extracted target scene image, in turn
The data set according to made of target scene image is facilitated to be trained, so that using trained prediction model to fire and smog
Prediction classification forecasting efficiency and prediction accuracy it is higher;Due to the meeting packet in the target scene image comprising fire and smog
Containing fire and smoke data, and different fire and smoke data can correspond to different scene types, such as when target scene figure
When there is bright flame as in, then open fire fire has occurred in explanation, and works as in target scene image and only exist strong smog
When, then smog alert has occurred in explanation, and very likely evolves into open fire fire, therefore extracts the fire in target scene image
Scene type is determined with smoke data and according to fire and smoke data, convenient for according to fire and smoke data and corresponding field
Scape classification makes data set, and is predicted convenient for the subsequent scene type to scene image to be measured to get prediction classification is arrived.
Further, in the step 11, the specific steps for obtaining the target scene image include:
Step 111: the scene image sample being normalized, mid-scene image is obtained;
Step 112: in the mid-scene image the fire and smoke data be labeled processing, obtain described
Target scene image;
The fire and smoke data include flame color feature data, flame texture characteristic, smog color characteristic
Data and smog textural characteristics data.
The beneficial effect of above-mentioned further scheme is: due to the type and burning of flame color and flame texture and comburant
The factors such as temperature it is related, therefore the fire under different scenes classification can have different flame color feature data and different
Flame texture characteristic, similarly the smog under different scenes classification can also have different smog color characteristic data and smog
Textural characteristics data, therefore by the way that scene image sample is normalized in advance, convenient for according to target prediction model
Support same substandard scene image sample (i.e. mid-scene image) be further processed, convenient for get fire and
Smoke data, by mid-scene image fire and smoke data be labeled processing, convenient for according to the mesh after mark
It marks scene image and extracts corresponding fire and smoke data, convenient for the prediction of the subsequent scene type to scene to be measured to get arriving
Predict classification.
Further, the step 2 specifically includes:
Step 21: obtaining the trained pre-training model in advance;
Step 22: the data set being divided into training set and test set, and using the training set to the pre-training mould
Type is trained, and obtains trained initial predicted model;
Step 23: training being iterated to the initial predicted model using the training set, obtains the target prediction
Model.
The beneficial effect of above-mentioned further scheme is: obtaining pre-training model, in advance convenient for using the method for transfer learning
Pre-training model is trained, convenient for obtaining target prediction model, and data set is divided into training set and test set, can contracted
Small training sample, and training time and training cost are reduced, and due to the randomness of division, training effect can be improved, be convenient for
Obtain the higher target prediction model of prediction accuracy;By repetitive exercise, further such that obtained target prediction model
Prediction accuracy is higher.
Further, the data set is specifically divided into the training set using arbitrary sampling method in the step 22
With the test set.
The beneficial effect of above-mentioned further scheme is: the division of data set can be made to have more using arbitrary sampling method
Randomness improves training effect.
Further, in the step 22, it is specific to use the method based on ResNet-152 model to the pre-training mould
Type is trained.
The beneficial effect of above-mentioned further scheme is: (being based on depth residual error network mould based on ResNet-152 model
Type) method is to add on the basis of traditional neural network (such as VGG model, Visual Geometry Group) in convolution interlayer
The mechanism of identical mapping is entered, so that several convolutional layers form a kind of packed structures, residual error unit has been formed, directly in receiving
One layer of output still keeps effect not lose to solve under the profound network architecture, can be in convolutional Neural
The network number of plies still keeps preferable performance in the case where deepening, it solves the net that traditional neural network occurs when the number of plies is deepened
The problem of network is degenerated;Therefore the method based on ResNet-152 model is used, it can be on the basis of being based on transfer learning method
The feature learning and characterization ability of pre-training model are inherited, and can solve the problem of network occurred when the number of plies is deepened is degenerated, is made
The prediction accuracy for the target prediction model that must be obtained is higher, and it is finally accurate to the prediction of the prediction classification of fire and smog to improve
Degree avoids the safety of life and property for threatening people to easily and timely find fire and smog alert.
Further, in the step 23, obtain further comprising the steps of after the target prediction model:
Step 24: test verifying is carried out to the target prediction model using the test set.
The beneficial effect of above-mentioned further scheme is: carrying out test verifying to target prediction model by test set, it is ensured that
The precision of prediction of obtained target prediction model, to guarantee timely and accurately to find fire and smog alert.
Another aspect according to the present invention provides a kind of fire based on transfer learning and smog forecasting system, including
Image acquisition unit, data set production unit, training unit and predicting unit;
Described image acquiring unit, for obtaining multiple scene image samples comprising fire and smog;
The data set production unit, for making data set according to the scene image sample;
The training unit, for being based on transfer learning method, using the data set to the pre-training mould obtained in advance
Type is trained, and obtains trained target prediction model;
The predicting unit is obtained for being predicted according to the target prediction model the scene image to be measured
The prediction classification of the scene image to be measured.
The beneficial effects of the present invention are: the fire and smog forecasting system of the invention based on transfer learning, passes through image
Acquiring unit, data set production unit and training unit obtain the higher target prediction model of precision of prediction, then single by prediction
Member predicts data set, obtains the prediction classification of the higher fire of precision of prediction and smog, is based on transfer learning, Neng Goushi
The demand of small sample training is answered, is not necessarily to a large amount of training sample, it will be able to obtain preferable learning effect, and the training time is significantly
Shorten, training cost substantially reduces, more conducively the feature extraction to fire and smog and representative learning, realizes to fire and smog
Prediction classification higher precision classifying quality, to improve target prediction model to the pre- of the prediction classification of fire and smog
Precision and forecasting efficiency are surveyed, to easily and timely find fire and smog alert, avoids the life for threatening people and property
Safety.
Another aspect according to the present invention provides a kind of fire based on transfer learning and smog forecasting system, including
Processor, memory and storage in the memory and may operate at computer program on the processor, the calculating
Machine program realizes the step in a kind of fire and smog prediction technique based on transfer learning of the invention when running.
The beneficial effects of the present invention are: the computer program by storage on a memory, and run on a processor, it is real
Existing fire and smog forecasting system of the invention based on transfer learning, can adapt to the demand of small sample training, without a large amount of
Training sample, it will be able to obtain preferable learning effect, and the training time greatly shortens, training cost substantially reduce, it is more sharp
In to fire and smog feature extraction and representative learning, realize the classification to the higher precision of the prediction classification of fire and smog
Effect, to improve target prediction model to the precision of prediction and forecasting efficiency of the prediction classification of fire and smog, thus convenient
Fire and smog alert are found in time, avoid the safety of life and property for threatening people.
Another aspect according to the present invention, provides a kind of computer storage medium, and the computer storage medium includes:
At least one instruction, is performed in described instruction and realizes a kind of fire and smog prediction side based on transfer learning of the invention
Step in method.
The beneficial effects of the present invention are: realizing this hair by executing the computer storage medium comprising at least one instruction
Bright fire and smog prediction based on transfer learning, can adapt to the demand of small sample training, without a large amount of training sample,
Preferable learning effect can be obtained, and the training time greatly shortens, training cost substantially reduces, more conducively to fire and cigarette
The feature extraction of mist and representative learning realize the classifying quality of the higher precision to the prediction classification of fire and smog, to mention
High target prediction model is to the precision of prediction and forecasting efficiency of the prediction classification of fire and smog, thus easily and timely discovery fire
Calamity and smog alert, avoid the safety of life and property for threatening people.
Detailed description of the invention
Fig. 1 is a kind of flow diagram one of fire and smog prediction technique based on transfer learning of the present invention;
Fig. 2 is a kind of flow diagram two of fire and smog prediction technique based on transfer learning of the present invention;
Fig. 3-1 is that the training loss of the prediction model not obtained using transfer learning method in the embodiment of the present invention one is bent
Line chart;
Fig. 3-2 is the training loss of the target prediction model obtained using transfer learning method in the embodiment of the present invention one
Curve graph;
Fig. 4-1 is that the precision of prediction of the prediction model not obtained using transfer learning method in the embodiment of the present invention one is bent
Line chart;
Fig. 4-2 is the precision of prediction of the target prediction model obtained using transfer learning method in the embodiment of the present invention one
Curve graph;
Fig. 5 is the fire and smog prediction result figure based on transfer learning in the embodiment of the present invention one;
Fig. 6 is a kind of structural schematic diagram of fire and smog forecasting system based on transfer learning of the present invention.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the invention.
With reference to the accompanying drawing, the present invention will be described.
Embodiment one, as shown in Figure 1, a kind of fire and smog prediction technique based on transfer learning, comprising the following steps:
S1: multiple scene image samples comprising fire and smog are obtained, and number is made according to the scene image sample
According to collection;
S2: it is based on transfer learning method, the pre-training model obtained in advance is trained using the data set, is obtained
Trained target prediction model;
S3: predicting scene image to be measured according to the target prediction model, obtains the scene image to be measured
Predict classification.
Since extraction and study of the convolutional neural networks model obtained in advance to feature can be inherited by transfer learning
Ability, therefore the application is based on transfer learning method, the data set made by scene image sample is to the pre- instruction obtained in advance
Practice model to be trained, the extraction of the feature of fire and smog that pre-training model concentrates data can be inherited and learns energy
Power can adapt to the demand of small sample training, be not necessarily to a large amount of training sample, it will be able to obtain preferable learning effect, and instruct
The white silk time greatly shortens, and training cost substantially reduces, more conducively the feature extraction to fire and smog and representative learning, realization pair
The classifying quality of the higher precision of the prediction classification of fire and smog, to improve target prediction model to the pre- of fire and smog
The precision of prediction and forecasting efficiency of classification are surveyed, to easily and timely find fire and smog alert, avoids threatening people's
Safety of life and property.
Preferably, as shown in Fig. 2, in S1, the specific steps for making the data set include:
S11: the scene image sample is pre-processed, target scene image is obtained;
S12: the fire and smoke data of the target scene image are extracted, and is determined according to the fire and smoke data
The corresponding scene type;
S13: the data set is made according to the fire and smoke data and the corresponding scene type.
By the pretreatment to scene image sample, convenient for the fire and smog number in subsequent extracted target scene image
According to, and then the data set according to made of target scene image is facilitated to be trained, so that using trained prediction model to fire
The forecasting efficiency and precision of prediction of calamity and the prediction classification of smog are higher;Due in the target scene image comprising fire and smog
Meeting include fire and smoke data, and different fire and smoke data can correspond to different scene types, such as work as target
When there is bright flame in scene image, then open fire alert has occurred in explanation, and strong when only existing in target scene image
Smog when, then smog alert has occurred in explanation, and very likely evolves into open fire fire, therefore extract in target scene image
Fire and smoke data and scene type is determined according to fire and smoke data, convenient for according to fire and smoke data and right
The scene type production data set answered, and predicted convenient for the subsequent scene type to scene image to be measured to get prediction is arrived
Classification.
Scene type correspondence is divided into three types by the present embodiment: open fire alert, smog alert and normal.
Preferably, as shown in Fig. 2, in S11, the specific steps for obtaining the target scene image include:
S111: the scene image sample is normalized, mid-scene image is obtained;
S112: in the mid-scene image the fire and smoke data be labeled processing, obtain the mesh
Mark scene image;
The fire and smoke data include flame color feature data, flame texture characteristic, smog color characteristic
Data and smog textural characteristics data.
Since flame color and flame texture are related with the factors such as the type of comburant and the temperature of burning, different fields
Fire under scape classification can have different flame color feature data and different flame texture characteristics, similarly different fields
Smog under scape classification can also have different smog color characteristic data and smog textural characteristics data, therefore by right in advance
Scene image sample is normalized, convenient for the same substandard scene image sample supported according to target prediction model
This (i.e. mid-scene image) is further processed, convenient for getting fire and smoke data, by mid-scene image
In fire and smoke data be labeled processing, convenient for according to the corresponding fire of target scene image zooming-out and cigarette after mark
Mist data, convenient for the subsequent scene type to scene image to be measured prediction to get to prediction classification.
Specifically, the normalization processing method selection of the present embodiment goes average value processing and image scaling to handle, can will be fiery
Calamity and smoke data feature normalization.
Preferably, as shown in Fig. 2, S2 is specifically included:
S21: the trained pre-training model is obtained in advance;
S22: being divided into training set and test set for the data set, and using the training set to the pre-training model into
Row training, obtains trained initial predicted model;
S23: training is iterated to the initial predicted model using the training set, obtains the target prediction mould
Type.
Pre-training model is obtained in advance, convenient for being trained using the method for transfer learning to pre-training model, convenient for
To target prediction model, and data set is divided into training set and test set, training sample can be reduced, and reduce the training time and
Training cost, and due to the randomness of division, training effect can be improved, convenient for obtaining the higher target prediction mould of precision of prediction
Type;By repetitive exercise, further such that the precision of prediction of obtained target prediction model is higher.
The pre-training model of the present embodiment can be the convolutional Neural net about fire and smog trained before oneself
Network Model Results are also possible to other people the convolutional neural networks Model Results about fire and smog.
As shown in Fig. 2, the data set is specifically divided into the training using arbitrary sampling method in the present embodiment S22
Collection and the test set, and specifically the pre-training model is trained using the method based on ResNet-152 model,
In, arbitrary sampling method specifically using without the methods of sampling put back to, finally uses the test set to the target prediction model
Test verifying is carried out, final target prediction model is obtained.
Method based on ResNet-152 model (depth residual error network) be traditional neural network (such as VGG model,
Visual Geometry Group) on the basis of, it joined the mechanism of identical mapping in convolution interlayer, so that several convolutional layers
A kind of packed structures are formed, residual error unit is formed, directly receives upper one layer of output, to solve in profound network
It still keeps effect not lose under framework, can still keep preferable property in the case where the convolutional neural networks number of plies is deepened
Can, it solves the problems, such as that the network that traditional neural network occurs when the number of plies is deepened is degenerated;Therefore using based on ResNet-
The method of 152 models can inherit the feature learning and characterization energy of pre-training model on the basis of based on transfer learning method
Power, and can solve the problem of network occurred when the number of plies is deepened is degenerated, so that the precision of prediction of obtained target prediction model is more
Height improves the precision of prediction of the finally prediction classification to fire and smog, thus easily and timely find fire and smog alert,
Avoid threatening the safety of life and property of people.
The present embodiment is obtained to the prediction model for transfer learning method not being used to obtain and using transfer learning method respectively
Target prediction model data set is predicted, by during prediction training loss and precision of prediction comparing respectively,
Obtained result is as shown in Table 1 and Table 2, and draws curve graph respectively, as shown in Fig. 3-1, Fig. 3-2, Fig. 4-1 and Fig. 4-2.
Table 1 is not using the prediction model of transfer learning method and the prediction model based on transfer learning method in identical iteration
The comparison of precision of prediction under number
Table 2 is not using the prediction model of transfer learning method and the prediction model based on transfer learning method in identical iteration
The comparison of training loss under number
From four width figure of Tables 1 and 2 and Fig. 3-1, Fig. 3-2, Fig. 4-1 and Fig. 4-2 as can be seen that in identical iteration time
Under several, the present invention is based on the fire of transfer learning and smog prediction technique to have smaller training loss and higher prediction essence
Degree, is highly suitable for the prediction of fire and smog, to easily and timely find fire and smog alert, avoids threatening people
Safety of life and property.
The present embodiment predicts that the result figure of obtained prediction classification is such as to the three width target scene images that data are concentrated
Shown in Fig. 5.
Embodiment two, as shown in fig. 6, a kind of fire and smog forecasting system based on transfer learning, including image obtains
Unit, data set production unit, training unit and predicting unit;
Described image acquiring unit, for obtaining multiple scene image samples comprising fire and smog;
The data set production unit, for making data set according to the scene image sample;
The training unit, for being based on transfer learning method, using the data set to the pre-training mould obtained in advance
Type is trained, and constructs trained target prediction model;
The predicting unit obtains described for being predicted according to the target prediction model scene image to be measured
The prediction classification of scene image to be measured.
Fire and smog forecasting system based on transfer learning of the invention is made by image acquisition unit, data set
Unit and training unit obtain the higher target prediction model of precision of prediction, then are predicted by predicting unit data set,
The prediction classification of the higher fire of precision of prediction and smog is obtained, transfer learning is based on, can adapt to the demand of small sample training,
Without a large amount of training sample, it will be able to obtain preferable learning effect, and the training time greatly shortens, training cost drops significantly
Low, the more conducively feature extraction to fire and smog and representative learning are realized to the more high-precision of the prediction classification of fire and smog
The classifying quality of degree, thus improve target prediction model to the precision of prediction and forecasting efficiency of the prediction classification of fire and smog,
To easily and timely find fire and smog alert, the safety of life and property for threatening people is avoided.
Embodiment three is based on embodiment one and embodiment two, and the present embodiment also discloses a kind of fire based on transfer learning
Calamity and smog forecasting system, including processor, memory and storage in the memory and may operate on the processor
Computer program, the computer program run when realize following steps as shown in Figure 1:
S1: multiple scene image samples comprising fire and smog are obtained, and number is made according to the scene image sample
According to collection;
S2: being based on transfer learning method, be trained using the data set to the pre-training model obtained in advance, is constructed
Trained target prediction model;
S3: predicting scene image to be measured according to the target prediction model, obtains the scene image to be measured
Predict classification.
It by storing computer program on a memory, and runs on a processor, realizes of the invention based on migration
The fire and smog forecasting system of study, can adapt to the demand of small sample training, be not necessarily to a large amount of training sample, it will be able to
To preferable learning effect, and the training time greatly shortens, and training cost substantially reduces, more conducively the feature to fire and smog
Extraction and representative learning realize the classifying quality of the higher precision to the prediction classification of fire and smog, so that it is pre- to improve target
Model is surveyed to the precision of prediction and forecasting efficiency of the prediction classification of fire and smog, to easily and timely find fire and smog
Alert avoids the safety of life and property for threatening people.
The present embodiment also provides a kind of computer storage medium, is stored at least one in the computer storage medium and refers to
It enables, described instruction is performed the specific steps for realizing the S1-S3.
By executing the computer storage medium comprising at least one instruction, the fire of the invention based on transfer learning is realized
Calamity and smog prediction, can adapt to the demand of small sample training, are not necessarily to a large amount of training sample, it will be able to preferably be learnt
Effect, and the training time greatly shortens, training cost substantially reduces, and the more conducively feature extraction to fire and smog and characterization is learned
It practises, the classifying quality of the higher precision to the prediction classification of fire and smog is realized, to improve target prediction model to fire
Prestige is avoided to easily and timely find fire and smog alert with the precision of prediction and forecasting efficiency of the prediction classification of smog
Coerce the safety of life and property of people.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of fire and smog prediction technique based on transfer learning, which comprises the following steps:
Step 1: obtaining multiple scene image samples comprising fire and smog, and data are made according to the scene image sample
Collection;
Step 2: being based on transfer learning method, the pre-training model obtained in advance is trained using the data set, is obtained
Target prediction model;
Step 3: scene image to be measured being predicted according to the target prediction model, obtains the pre- of the scene image to be measured
Survey classification.
2. the fire and smog prediction technique according to claim 1 based on transfer learning, which is characterized in that in the step
In rapid 1, the specific steps for making the data set include:
Step 11: the scene image sample being pre-processed, target scene image is obtained;
Step 12: extracting the fire and smoke data of the target scene image, and determined according to the fire and smoke data
Corresponding scene type;
Step 13: the data set is made according to the fire and smoke data and the corresponding scene type.
3. the fire and smog prediction technique according to claim 2 based on transfer learning, which is characterized in that in the step
In rapid 11, the specific steps for obtaining the target scene image include:
Step 111: the scene image sample being normalized, mid-scene image is obtained;
Step 112: in the mid-scene image the fire and smoke data be labeled processing, obtain the target
Scene image;
The fire and smoke data include flame color feature data, flame texture characteristic, smog color characteristic data
With smog textural characteristics data.
4. the fire and smog prediction technique according to claim 1 based on transfer learning, which is characterized in that the step
2 specifically include:
Step 21: obtaining the trained pre-training model in advance;
Step 22: the data set is divided into training set and test set, and using the training set to the pre-training model into
Row training, obtains trained initial predicted model;
Step 23: training being iterated to the initial predicted model using the training set, obtains the target prediction model.
5. the fire and smog prediction technique according to claim 4 based on transfer learning, which is characterized in that in the step
In rapid 22, the data set is specifically divided by the training set and the test set using arbitrary sampling method.
6. the fire and smog prediction technique according to claim 4 based on transfer learning, which is characterized in that in the step
It is specific that the pre-training model is trained using the method based on ResNet-152 model in rapid 22.
7. the fire and smog prediction technique according to claim 4 based on transfer learning, which is characterized in that in the step
In rapid 23, obtain further comprising the steps of after the target prediction model:
Step 24: test verifying is carried out to the target prediction model using the test set.
8. a kind of fire and smog forecasting system based on transfer learning, which is characterized in that including image acquisition unit, data set
Production unit, training unit and predicting unit;
Described image acquiring unit, for obtaining multiple scene image samples comprising fire and smog;
The data set production unit, for making data set according to the scene image sample;
The training unit, for be based on transfer learning method, using the data set to the pre-training model obtained in advance into
Row training, obtains target prediction model;
The predicting unit obtains described to be measured for being predicted according to the target prediction model scene image to be measured
The prediction classification of scene image.
9. a kind of fire and smog forecasting system based on transfer learning, which is characterized in that including processor, memory and storage
It in the memory and may operate at the computer program on the processor, realize when the computer program is run as weighed
Benefit requires method and step described in any one of 1-7 claim.
10. a kind of computer storage medium, which is characterized in that the computer storage medium includes: at least one instruction, in institute
It states instruction and is performed realization the method according to claim 1 to 7 step.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811583508.3A CN109858516A (en) | 2018-12-24 | 2018-12-24 | A kind of fire and smog prediction technique, system and medium based on transfer learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811583508.3A CN109858516A (en) | 2018-12-24 | 2018-12-24 | A kind of fire and smog prediction technique, system and medium based on transfer learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109858516A true CN109858516A (en) | 2019-06-07 |
Family
ID=66892135
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811583508.3A Pending CN109858516A (en) | 2018-12-24 | 2018-12-24 | A kind of fire and smog prediction technique, system and medium based on transfer learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109858516A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110443197A (en) * | 2019-08-05 | 2019-11-12 | 珠海格力电器股份有限公司 | A kind of visual scene intelligent Understanding method and system |
CN110765937A (en) * | 2019-10-22 | 2020-02-07 | 新疆天业(集团)有限公司 | Coal yard spontaneous combustion detection method based on transfer learning |
CN111414514A (en) * | 2020-03-19 | 2020-07-14 | 山东雷火网络科技有限公司 | System and method for flame detection based on Shandong Jinnan province |
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 |
CN112349057A (en) * | 2020-12-01 | 2021-02-09 | 北京交通大学 | Deep learning-based indoor smoke and fire detection method |
CN112529022A (en) * | 2019-08-28 | 2021-03-19 | 杭州海康威视数字技术股份有限公司 | Training sample generation method and device |
CN112598071A (en) * | 2020-12-28 | 2021-04-02 | 北京市商汤科技开发有限公司 | Open fire identification method, device, equipment and storage medium |
CN112699963A (en) * | 2021-01-13 | 2021-04-23 | 四川九通智路科技有限公司 | Fire detection method |
CN113128412A (en) * | 2021-04-22 | 2021-07-16 | 重庆大学 | Fire trend prediction method based on deep learning and fire monitoring video |
CN114371145A (en) * | 2022-03-21 | 2022-04-19 | 武汉工程大学 | Detection method and device for milk oil mixed pigment, electronic equipment and storage medium |
CN116563673A (en) * | 2023-07-10 | 2023-08-08 | 浙江华诺康科技有限公司 | Smoke training data generation method and device and computer equipment |
CN117409529A (en) * | 2023-10-13 | 2024-01-16 | 国网江苏省电力有限公司南通供电分公司 | Multi-scene electrical fire on-line monitoring method and system |
-
2018
- 2018-12-24 CN CN201811583508.3A patent/CN109858516A/en active Pending
Non-Patent Citations (1)
Title |
---|
王文朋 等: "基于深度迁移学习的烟雾识别方法", 《计算机应用》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110443197A (en) * | 2019-08-05 | 2019-11-12 | 珠海格力电器股份有限公司 | A kind of visual scene intelligent Understanding method and system |
CN112529022A (en) * | 2019-08-28 | 2021-03-19 | 杭州海康威视数字技术股份有限公司 | Training sample generation method and device |
CN112529022B (en) * | 2019-08-28 | 2024-03-01 | 杭州海康威视数字技术股份有限公司 | Training sample generation method and device |
CN110765937A (en) * | 2019-10-22 | 2020-02-07 | 新疆天业(集团)有限公司 | Coal yard spontaneous combustion detection method based on transfer learning |
CN111414514B (en) * | 2020-03-19 | 2024-01-19 | 山东雷火网络科技有限公司 | System and method for flame detection in Shandong Jinan environment |
CN111414514A (en) * | 2020-03-19 | 2020-07-14 | 山东雷火网络科技有限公司 | System and method for flame detection based on Shandong Jinnan province |
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 |
CN112349057A (en) * | 2020-12-01 | 2021-02-09 | 北京交通大学 | Deep learning-based indoor smoke and fire detection method |
CN112598071A (en) * | 2020-12-28 | 2021-04-02 | 北京市商汤科技开发有限公司 | Open fire identification method, device, equipment and storage medium |
CN112699963A (en) * | 2021-01-13 | 2021-04-23 | 四川九通智路科技有限公司 | Fire detection method |
CN113128412A (en) * | 2021-04-22 | 2021-07-16 | 重庆大学 | Fire trend prediction method based on deep learning and fire monitoring video |
CN113128412B (en) * | 2021-04-22 | 2022-06-07 | 重庆大学 | Fire trend prediction method based on deep learning and fire monitoring video |
CN114371145A (en) * | 2022-03-21 | 2022-04-19 | 武汉工程大学 | Detection method and device for milk oil mixed pigment, electronic equipment and storage medium |
CN116563673B (en) * | 2023-07-10 | 2023-12-12 | 浙江华诺康科技有限公司 | Smoke training data generation method and device and computer equipment |
CN116563673A (en) * | 2023-07-10 | 2023-08-08 | 浙江华诺康科技有限公司 | Smoke training data generation method and device and computer equipment |
CN117409529A (en) * | 2023-10-13 | 2024-01-16 | 国网江苏省电力有限公司南通供电分公司 | Multi-scene electrical fire on-line monitoring method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109858516A (en) | A kind of fire and smog prediction technique, system and medium based on transfer learning | |
CN107633255B (en) | Rock lithology automatic identification and classification method under deep learning mode | |
CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
CN107730000A (en) | Coolant leakage detection method and device for air-conditioning | |
CN106844614A (en) | A kind of floor plan functional area system for rapidly identifying | |
CN108090472B (en) | Pedestrian re-identification method and system based on multi-channel consistency characteristics | |
CN109543679A (en) | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks | |
CN111079784B (en) | Flue-cured tobacco baking stage identification method in baking process based on convolutional neural network | |
CN110826453A (en) | Behavior identification method by extracting coordinates of human body joint points | |
CN109325495A (en) | A kind of crop image segmentation system and method based on deep neural network modeling | |
CN109508671A (en) | A kind of video accident detection system and method based on Weakly supervised study | |
CN106503458A (en) | A kind of surface air temperature data quality control method | |
CN112115993B (en) | Zero sample and small sample evidence photo anomaly detection method based on meta-learning | |
EP3467712B1 (en) | Methods and systems for processing image data | |
Shi et al. | Amur tiger stripes: Individual identification based on deep convolutional neural network | |
CN109284735B (en) | Mouse feelings monitoring method, device and storage medium | |
CN112966714A (en) | Edge time sequence data anomaly detection and network programmable control method | |
CN117169150A (en) | Method for detecting total bacteria on surface of cooled mutton | |
CN113065601A (en) | Deep learning forest fire abnormity detection method based on genetic algorithm optimization | |
CN113033547A (en) | Welding state classification method based on MobileNet V2 | |
CN109670423A (en) | A kind of image identification system based on deep learning, method and medium | |
Tian et al. | A method of corn disease identification based on convolutional neural network | |
CN111062350A (en) | Artificial intelligence based firework identification algorithm | |
CN109034033B (en) | Smoke discharge video detection method based on improved VGG16 convolutional network | |
CN110045691A (en) | A kind of multitasking fault monitoring method of multi-source heterogeneous big data |
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 |