CN110348505A - Vehicle color disaggregated model training method, device and vehicle color identification method - Google Patents
Vehicle color disaggregated model training method, device and vehicle color identification method Download PDFInfo
- Publication number
- CN110348505A CN110348505A CN201910590409.6A CN201910590409A CN110348505A CN 110348505 A CN110348505 A CN 110348505A CN 201910590409 A CN201910590409 A CN 201910590409A CN 110348505 A CN110348505 A CN 110348505A
- Authority
- CN
- China
- Prior art keywords
- vehicle color
- vehicle
- disaggregated model
- color
- feature set
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle color disaggregated model training methods, comprising: several vehicle image data that will acquire are input in preparatory trained deep neural network, to extract vehicle color feature set；Cluster operation is carried out to vehicle color feature set using clustering algorithm, obtains the subset under several different scenes classifications；Vehicle image under scene type corresponding with subset is input in corresponding vehicle color classification submodel, to be trained to vehicle color classification submodel；The output layer of all vehicle colors classification submodel is linked to the input layer of vehicle color disaggregated model, to train vehicle color disaggregated model；Wherein, vehicle color classification submodel and vehicle color disaggregated model are designed to obtain by deep neural network.Invention additionally discloses a kind of vehicle color disaggregated model training device and a kind of vehicle color identification methods.Using the embodiment of the present invention, the variation of different scenes under body color can be coped with, improves vehicle color accuracy of identification.
Description
Technical field
The present invention relates to image identification technical field more particularly to a kind of vehicle color disaggregated model training methods, device
And vehicle color identification method.
Background technique
Body color identification is the critical issue in vehicle attribute identification field, is the base for carrying out vehicle feature recognition and analysis
Plinth has important researching value.With the popularization and use of deep learning theory, vehicle detection accuracy rate is constantly promoted, base
Testing result in the positions such as vehicle body or vehicle face, existing body color recognition methods are concentrated mainly on following mode:
The research that mode one, color detection regions position.The premise for carrying out vehicle color identification is that one piece of acquisition is stable
Color identification region, such as using the segment rectangle on front cover for vehicle as color detection regions, or use the conduct of vehicle region
Color detection regions.
Mode two, the research based on vehicle image segmentation algorithm.In addition to the method for position color identification region, vehicle segmentation
It is also the main method for carrying out vehicle color identification.Using image processing algorithm, body portion and background parts are split,
To carry out color identification for body portion.
Vehicle color identification under complex scene is the research hotspot in vehicle attribute identification field, although achieving certain
Achievement, but still remain many technological difficulties: vehicle color is influenced vulnerable to the scenes factor such as illumination, dust, and color itself is
One has the feature of subjectivity, influence of the vehicle color due to the scenes factor such as illumination, dust in monitoring scene, human eye
Also the case where being often difficult to.Color detection regions are difficult to be accurately positioned, under outdoor complex scene, either base
It, all can be by the posture, multiple of current detection algorithm, vehicle during interception in the color identification region of vehicle body or vehicle face
The influence of miscellaneous background interference or foreground occlusion, the vehicle color identification region for intercepting out is often with a large amount of non-vehicle face
The information of color.
Summary of the invention
Know the purpose of the embodiment of the present invention is that providing a kind of vehicle color disaggregated model training method, device and vehicle color
Other method can cope with the variation of different scenes under body color, improve vehicle color accuracy of identification.
To achieve the above object, the embodiment of the invention provides a kind of vehicle color disaggregated model training methods, comprising:
Several vehicle images are input in preparatory trained deep neural network, to extract vehicle color feature
Collection；
Cluster operation is carried out to the vehicle color feature set using clustering algorithm, is obtained under several different scenes classifications
Subset；
Vehicle image under scene type corresponding with the subset is input to corresponding vehicle color classification submodel
In, to be trained to vehicle color classification submodel；
The output layer of all vehicle color classification submodels is linked to the input layer of vehicle color disaggregated model, with
The training vehicle color disaggregated model；Wherein, the vehicle color classification submodel and the vehicle color disaggregated model are equal
It designs to obtain by deep neural network.
Compared with prior art, vehicle color disaggregated model training method disclosed by the invention, firstly, vehicle image is defeated
Enter into preparatory trained deep neural network, image data is converted into numeric type data by neural depth network, is mentioned
Take out vehicle color feature set；Then, cluster operation is carried out to the vehicle color feature set using clustering algorithm, it can basis
The scene similitude of image divides data set to several corresponding subsets；Finally, corresponding vehicle is respectively trained according to the subset
Color classification submodel, and the output layer of all vehicle colors classification submodels is linked to vehicle color disaggregated model
Input layer, with the training vehicle color disaggregated model.Vehicle color classification submodel under all scenes is integrated
Trained vehicle color disaggregated model has stronger adaptability to each scene illumination, when can cope with illumination condition variation
The variation of body color, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
It is as an improvement of the above scheme, described that cluster operation is carried out to the vehicle color feature set using clustering algorithm,
The subset under several different scenes classifications is obtained, formula is met:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckFor
The number of the subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
As an improvement of the above scheme, the clustering algorithm is K-Means algorithm.
As an improvement of the above scheme, the output layer by all vehicle color classification submodels is linked to vehicle
Before the input layer of color classification model, further includes:
Delete the full articulamentum and classification layer in the vehicle color classification submodel.
As an improvement of the above scheme, the deep neural network include in VGG16, ResNet, SqueezeNet extremely
Few one kind.
To achieve the above object, the embodiment of the invention also provides a kind of vehicle color disaggregated model training devices, comprising:
Vehicle color feature set extraction module, for several vehicle images to be input to trained depth nerve in advance
In network, to extract vehicle color feature set；
Computing module is clustered, for carrying out cluster operation to the vehicle color feature set using clustering algorithm, if obtaining
Subset under dry different scenes classification；
Vehicle color is classified submodel training module, for by the vehicle image under scene type corresponding with the subset
It is input in corresponding vehicle color classification submodel, to be trained to vehicle color classification submodel；
Vehicle color disaggregated model training module, for linking the output layer of all vehicle color classification submodels
To the input layer of vehicle color disaggregated model, with the training vehicle color disaggregated model；Wherein, vehicle color classification
Model and the vehicle color disaggregated model are designed to obtain by deep neural network.
Compared with prior art, vehicle color disaggregated model training device disclosed by the invention, firstly, vehicle color feature
Vehicle image is input in preparatory trained deep neural network by collection extraction module, by neural depth network by picture number
According to numeric type data is converted to, vehicle color feature set is extracted；Then, cluster computing module is using clustering algorithm to the vehicle
Color characteristic collection carries out cluster operation, can be according to the scene similitude of image by several corresponding subsets of data set point；
Finally, vehicle color is classified, corresponding vehicle color classification submodel is respectively trained according to the subset in submodel training module,
The output layer of all vehicle color classification submodels is linked to vehicle color point by vehicle color disaggregated model training module
The input layer of class model, with the training vehicle color disaggregated model.Vehicle color classification submodel under all scenes is comprehensive
Trained vehicle color disaggregated model altogether, has stronger adaptability to each scene illumination, can cope with illumination condition
The variation of body color when variation, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
It is as an improvement of the above scheme, described that cluster operation is carried out to the vehicle color feature set using clustering algorithm,
The subset under several different scenes classifications is obtained, formula is met:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckFor
The number of the subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
As an improvement of the above scheme, the clustering algorithm is K-Means algorithm.
As an improvement of the above scheme, described device further includes vehicle color classification submodel processing module, the vehicle
Color classification submodel processing module is used to delete full articulamentum and classification layer in the vehicle color classification submodel.
To achieve the above object, the embodiment of the present invention also provides a kind of vehicle color identification method, comprising:
Obtain the images to be recognized of vehicle；
The images to be recognized is inputted in trained vehicle color disaggregated model in advance；Wherein, the vehicle color
The training method of disaggregated model is such as vehicle color disaggregated model training method according to any one of claims 1 to 5；
Obtain the result that the output data of the vehicle color disaggregated model is identified as the vehicle color.
Detailed description of the invention
Fig. 1 is a kind of flow chart of vehicle color disaggregated model training method provided in an embodiment of the present invention；
Fig. 2 is vehicle color classification submodule in a kind of vehicle color disaggregated model training method provided in an embodiment of the present invention
The training frame of type；
Fig. 3 is vehicle color disaggregated model in a kind of vehicle color disaggregated model training method provided in an embodiment of the present invention
Training frame；
Fig. 4 is a kind of structural schematic diagram of vehicle color disaggregated model training device provided in an embodiment of the present invention；
Fig. 5 is a kind of flow chart of vehicle color identification method provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
It is worth noting that often have thousands of a cameras in video monitoring system, the deployment ring of these cameras
Border can not be completely the same, and therefore, the image data from these cameras also has many difference, and wherein illumination is most critical
Influence factor.Although the complicated multiplicity of the photographed scene of image, it still can find the regularity of distribution therein.Same
Period, the camera in same place, due to different illumination conditions, the image of shooting may have very big difference；Without
More consistent picture quality may also be presented because of similar illumination condition in the camera of same period, different location.Cause
This, is trained in the embodiment of the present invention according to obtaining corresponding vehicle image under different illumination scenes, can to train
Suitable for the vehicle color disaggregated model under several scenes classification.
Embodiment one
It is a kind of flow chart of vehicle color disaggregated model training method provided in an embodiment of the present invention referring to Fig. 1, Fig. 1；
Include:
S11, several vehicle images are input in preparatory trained deep neural network, to extract vehicle color
Feature set；
S12, cluster operation is carried out to the vehicle color feature set using clustering algorithm, obtains several different scenes classes
Subset under not；
S13, the vehicle image under scene type corresponding with the subset is input to corresponding vehicle color classification
In model, to be trained to vehicle color classification submodel；
S14, the input that the output layer of all vehicle color classification submodels is linked to vehicle color disaggregated model
Layer, with the training vehicle color disaggregated model；Wherein, the vehicle color classification submodel and vehicle color classification mould
Type is designed to obtain by deep neural network.
Specifically, in step s 11, image data can not directly carry out cluster operation, it is therefore necessary to by depth
Neural network is converted into numeric type feature.Preferably, the deep neural network include VGG16, ResNet,
At least one of SqueezeNet.It, can be to depth nerve due to the shallow-layer feature that color characteristic is image, therefore in extraction process
Network does appropriate cutting.If vehicle color data set is V (several vehicle images inputted, the vehicle color data set packet
Contain the vehicle image of different scenes classification, different colours), it is D by the vehicle color feature set that neural network is extracted, by vehicle
Image is converted to numeric type data.
Specifically, in step s 12, for vehicle color feature set D, if xi∈ D, i=1 ... N is a vehicle image
Numerical characteristics, D is clustered using K-Means algorithm, it is therefore an objective to the image of all similar scenes is clustered, meet
Formula:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckFor
The number of the subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
It is clustered, data set is divided into according to scene several by numerical characteristics of the K-Means to vehicle color image
The class internal difference mutation of subset, each subset is small, reduces performance concussion for the deep neural network model of each trained,
Nicety of grading can achieve higher level under the scene.
Preferably, the scene type can be divided into micro- exposure scene, balanced light field scape, shade light field scape, low-light field
Scape and no light scene, the scene type is divided into above-mentioned five kinds of classifications in embodiments of the present invention, but in other embodiments,
The scene type can be set according to the actual situation, it is not limited to above-mentioned five kinds of scenes, all in protection scope of the present invention
It is interior.By clustering operation, data set can be divided by several subsets according to the scene similitude of image, the figure in each subset
As may be from different cameras, it is also possible to which from the different periods, but they have similar scene and illumination
Condition, therefore there is similitude.
Specifically, in step s 13, for the subset Sj, it separately designs corresponding deep neural network and is trained,
The data inputted in the training process are that presently described subset corresponds to the vehicle image under scene type, to obtain k correspondence
Vehicle color classify submodel, herein by the vehicle color classification submodel Uniform Name be Teacher model.It is worth saying
Bright, each Teacher model obtained at this time can individually identify vehicle image, such as when described
When Teacher model is the vehicle color classification submodel under micro- exposure scene, the vehicle figure under micro- exposure scene is inputted at this time
Picture can carry out color identification to vehicle image.Vehicle color is identified for different scenes, can be avoided because not sharing the same light
The problem of causing vehicle color to be difficult to according under.
As shown in Fig. 2, Fig. 2 is vehicle face in a kind of vehicle color disaggregated model training method provided in an embodiment of the present invention
The training frame of colour sorting submodel；Wherein, the corresponding scene type of Teacher1 is micro- exposure scene, and Teacher2 is corresponding
Scene type is balanced light field scape, and the corresponding scene type of Teacher is shade light field scape, the corresponding scene type of Teacher4
For low-light scene, the corresponding scene type of Teacher5 is no light scene.Since the scene of the subset, illumination have one
Fixed similitude, therefore each Teacher model can carry out the higher body color classification of accuracy for its training subset.
Specifically, based on k obtained teacher model, new deep neural network model is designed in step S14,
The i.e. described vehicle color disaggregated model, is named as Student for the vehicle color disaggregated model herein.Utilize k Teacher
Model guides training to Student model, and the weight parameter of Teacher model no longer updates at this time, has only learned it
For the scene knowledge migration practised into Student, the specific vehicle color disaggregated model frame is as shown in Figure 3.
In training student model, the last full articulamentum of 5 teacher models and softmax classification layer are deleted
Fall, and the input layer that the output layer of all teacher models is linked to student model was so being trained
Cheng Zhong, teacher model will input respective priori knowledge to student respectively, by each teacher model to self-field
The ability in feature extraction of scape moves in student model, so that student model adapts to multiple scenes, can cope with illumination
The variation of body color when condition changes, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
Compared with prior art, vehicle color disaggregated model training method disclosed by the invention, firstly, vehicle image is defeated
Enter into preparatory trained deep neural network, image data is converted into numeric type data by neural depth network, is mentioned
Take out vehicle color feature set；Then, cluster operation is carried out to the vehicle color feature set using clustering algorithm, it can basis
The scene similitude of image divides data set to several corresponding subsets；Finally, corresponding vehicle is respectively trained according to the subset
Color classification submodel, and the output layer of all vehicle colors classification submodels is linked to vehicle color disaggregated model
Input layer, with the training vehicle color disaggregated model.Vehicle color classification submodel under all scenes is integrated
Trained vehicle color disaggregated model has stronger adaptability to each scene illumination, when can cope with illumination condition variation
The variation of body color, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
Embodiment two
Referring to fig. 4, Fig. 4 is a kind of structural representation of vehicle color disaggregated model training device provided in an embodiment of the present invention
Figure；Include:
Vehicle color feature set extraction module 11, for several vehicle images to be input to trained depth mind in advance
Through in network, to extract vehicle color feature set；
Cluster computing module 12 is obtained for carrying out cluster operation to the vehicle color feature set using clustering algorithm
Subset under several different scenes classifications；
Vehicle color is classified submodel training module 13, for by the vehicle figure under scene type corresponding with the subset
As being input in corresponding vehicle color classification submodel, to be trained to vehicle color classification submodel；
Vehicle color disaggregated model training module 14, for by the output layer chain of all vehicle colors classification submodels
It is connected to the input layer of vehicle color disaggregated model, with the training vehicle color disaggregated model；Wherein, the vehicle color classification
Submodel and the vehicle color disaggregated model are designed to obtain by deep neural network；
Vehicle color classification submodel processing module 15, for deleting the full connection in the vehicle color classification submodel
Layer and classification layer.
Specifically, image data can not directly carry out cluster operation, it is therefore necessary to by deep neural network by its
Be converted to numeric type feature.Preferably, the deep neural network includes at least one in VGG16, ResNet, SqueezeNet
Kind.Due to the shallow-layer feature that color characteristic is image, therefore in extraction process, appropriate cutting can be done to deep neural network.If vehicle
Color data integrates that (several vehicle images inputted, the vehicle color data set contain different scenes classification, no as V
With the vehicle image of color), the vehicle color feature that the vehicle color feature set extraction module 11 is extracted by neural network
Integrate as D, image data is converted into numeric type data.
Specifically, vehicle color feature set D is directed to, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image, institute
It states cluster computing module 12 to cluster D using K-Means algorithm, it is therefore an objective to gather the image of all similar scenes
Class meets formula:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckFor
The number of the subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
It is clustered, data set is divided into according to scene several by numerical characteristics of the K-Means to vehicle color image
The class internal difference mutation of subset, each subset is small, reduces performance concussion for the deep neural network model of each trained,
Nicety of grading can achieve higher level under the scene.
Preferably, the scene type can be divided into micro- exposure scene, balanced light field scape, shade light field scape, low-light field
Scape and no light scene, the scene type is divided into above-mentioned five kinds of classifications in embodiments of the present invention, but in other embodiments,
The scene type can be set according to the actual situation, it is not limited to above-mentioned five kinds of scenes, all in protection scope of the present invention
It is interior.By clustering operation, data set can be divided by a subset according to the scene similitude of image, the image in each subset can
It can be from different cameras, it is also possible to from the different periods, but they have similar scene and illumination condition,
Therefore there is similitude.
Specifically, the vehicle color classification submodel training module 13 is directed to the subset Sj, separately design corresponding
Deep neural network is trained, and the data inputted in the training process are that presently described subset corresponds to the vehicle under scene type
Image, so that k corresponding vehicle color classification submodels are obtained, herein by vehicle color classification submodel Uniform Name
For Teacher model.
Specifically, new deep neural network model is designed based on k obtained teacher model, i.e., the described vehicle face
The vehicle color disaggregated model is named as Student herein by colour sorting model.The vehicle color disaggregated model training mould
Block 14 guides training to Student model using k Teacher model, and the weight parameter of Teacher model is no longer at this time
It updates, the scene knowledge migration for only having learnt it is into Student.
In training student model, vehicle color classification submodel processing module 15 is by 5 teacher models
Last full articulamentum and softmax classification layer delete, and the vehicle color disaggregated model training module 14 will be all described
The output layer of teacher model is linked to the input layer of student model, so, in the training process, teacher mould
Type will input respective priori knowledge to student respectively, by each teacher model to the feature extraction energy of itself scene
Power moves in student model, so that student model adapts to multiple scenes, vehicle body when can cope with illumination condition variation
The variation of color, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
Compared with prior art, vehicle color disaggregated model training device disclosed by the invention, firstly, vehicle color feature
Vehicle image is input in preparatory trained deep neural network by collection extraction module 11, by neural depth network by image
Data are converted to numeric type data, extract vehicle color feature set；Then, cluster computing module 12 is using clustering algorithm to institute
It states vehicle color feature set and carries out cluster operation, data set can be divided to several corresponding sons according to the scene similitude of image
Collection；Finally, vehicle color is classified, corresponding vehicle color classification is respectively trained according to the subset in submodel training module 13
The output layer of all vehicle color classification submodels is linked to vehicle by model, vehicle color disaggregated model training module 14
The input layer of color classification model, with the training vehicle color disaggregated model.Vehicle color under all scenes is classified sub
Model integrates trained vehicle color disaggregated model, has stronger adaptability to each scene illumination, can cope with light
The variation of body color when according to condition variation, the adaptability and robustness of model are stronger, reach higher color accuracy of identification.
Embodiment three
It is a kind of flow chart of vehicle color identification method provided in an embodiment of the present invention referring to Fig. 5, Fig. 5；Include:
S21, the images to be recognized for obtaining vehicle；
S22, the images to be recognized is inputted in trained vehicle color disaggregated model in advance；Wherein, the vehicle
The training method of color classification model is the vehicle color disaggregated model training method as described in above-described embodiment one；
The result that S23, the output data for obtaining the vehicle color disaggregated model are identified as the vehicle color.
Vehicle color identification method described in the embodiment of the present invention uses vehicle color disaggregated model described in embodiment one
Vehicle color disaggregated model training pattern in training method after training, can cope with the variation of different scenes under body color, mention
High vehicle color accuracy of identification.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (10)
1. a kind of vehicle color disaggregated model training method characterized by comprising
Several vehicle images are input in preparatory trained deep neural network, to extract vehicle color feature set；
Cluster operation is carried out to the vehicle color feature set using clustering algorithm, obtains the son under several different scenes classifications
Collection；
Vehicle image under scene type corresponding with the subset is input in corresponding vehicle color classification submodel, with
Vehicle color classification submodel is trained；
The output layer of all vehicle color classification submodels is linked to the input layer of vehicle color disaggregated model, with training
The vehicle color disaggregated model；Wherein, the vehicle color classification submodel and the vehicle color disaggregated model are by depth
Degree neural network designs to obtain.
2. vehicle color disaggregated model training method as described in claim 1, which is characterized in that described to utilize clustering algorithm pair
The vehicle color feature set carries out cluster operation, obtains the subset under several different scenes classifications, meets formula:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckIt is described
The number of subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
3. vehicle color disaggregated model training method as claimed in claim 2, which is characterized in that the clustering algorithm is K-
Means algorithm.
4. vehicle color disaggregated model training method as described in claim 1, which is characterized in that described by all vehicles
Before the output layer of color classification submodel is linked to the input layer of vehicle color disaggregated model, further includes:
Delete the full articulamentum and classification layer in the vehicle color classification submodel.
5. vehicle color disaggregated model training method as described in claim 1, which is characterized in that the deep neural network packet
Include at least one of VGG16, ResNet, SqueezeNet.
6. a kind of vehicle color disaggregated model training device characterized by comprising
Vehicle color feature set extraction module, for several vehicle images to be input to preparatory trained deep neural network
In, to extract vehicle color feature set；
It clusters computing module and obtains several for carrying out cluster operation to the vehicle color feature set using clustering algorithm
Subset under different scenes classification；
Vehicle color classification submodel training module, for inputting the vehicle image under scene type corresponding with the subset
Into corresponding vehicle color classification submodel, to be trained to vehicle color classification submodel；
Vehicle color disaggregated model training module, for the output layer of all vehicle color classification submodels to be linked to vehicle
The input layer of color classification model, with the training vehicle color disaggregated model；Wherein, the vehicle color classification submodel
It designs to obtain by deep neural network with the vehicle color disaggregated model.
7. vehicle color disaggregated model training device as claimed in claim 6, which is characterized in that described to utilize clustering algorithm pair
The vehicle color feature set carries out cluster operation, obtains the subset under several different scenes classifications, meets formula:
Wherein, for vehicle color feature set D, if xi∈ D, i=1 ... N is the numerical characteristics of a vehicle image；ckIt is described
The number of subset；The vehicle color feature set is divided into several subset Sj,j∈1,…k。
8. vehicle color disaggregated model training device as claimed in claim 7, which is characterized in that the clustering algorithm is K-
Means algorithm.
9. vehicle color disaggregated model training device as claimed in claim 7, which is characterized in that described device further includes vehicle
Color classification submodel processing module, the vehicle color classification submodel processing module is for deleting the vehicle color classification
Full articulamentum and classification layer in submodel.
10. a kind of vehicle color identification method characterized by comprising
Obtain the images to be recognized of vehicle；
The images to be recognized is inputted in trained vehicle color disaggregated model in advance；Wherein, the vehicle color classification
The training method of model is such as vehicle color disaggregated model training method according to any one of claims 1 to 5；
Obtain the result that the output data of the vehicle color disaggregated model is identified as the vehicle color.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910590409.6A CN110348505B (en) | 2019-07-02 | 2019-07-02 | Vehicle color classification model training method and device and vehicle color identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910590409.6A CN110348505B (en) | 2019-07-02 | 2019-07-02 | Vehicle color classification model training method and device and vehicle color identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110348505A true CN110348505A (en) | 2019-10-18 |
| CN110348505B CN110348505B (en) | 2022-02-22 |
Family
ID=68177442
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910590409.6A Active CN110348505B (en) | 2019-07-02 | 2019-07-02 | Vehicle color classification model training method and device and vehicle color identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110348505B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126133A (en) * | 2019-11-08 | 2020-05-08 | 博云视觉（北京）科技有限公司 | Intelligent refrigerator access action recognition method based on deep learning |
| CN111291812A (en) * | 2020-02-11 | 2020-06-16 | 浙江大华技术股份有限公司 | Attribute class acquisition method and device, storage medium and electronic device |
| CN112016433A (en) * | 2020-08-24 | 2020-12-01 | 高新兴科技集团股份有限公司 | Vehicle color identification method based on deep neural network |
| CN116563770A (en) * | 2023-07-10 | 2023-08-08 | 四川弘和数智集团有限公司 | Method, device, equipment and medium for detecting vehicle color |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140063249A1 (en) * | 2012-09-05 | 2014-03-06 | Motorola Solutions, Inc. | Method, apparatus and system for performing facial recognition |
| CN103996041A (en) * | 2014-05-15 | 2014-08-20 | 武汉睿智视讯科技有限公司 | Vehicle color identification method and system based on matching |
| CN106778583A (en) * | 2016-12-07 | 2017-05-31 | 北京理工大学 | Vehicle attribute recognition methods and device based on convolutional neural networks |
| CN107067011A (en) * | 2017-03-20 | 2017-08-18 | 北京邮电大学 | A kind of vehicle color identification method and device based on deep learning |
| CN108875754A (en) * | 2018-05-07 | 2018-11-23 | 华侨大学 | A kind of vehicle recognition methods again based on more depth characteristic converged network |
| CN109376781A (en) * | 2018-10-24 | 2019-02-22 | 深圳市腾讯网络信息技术有限公司 | A kind of training method, image-recognizing method and the relevant apparatus of image recognition model |
| CN109558872A (en) * | 2018-11-22 | 2019-04-02 | 四川大学 | A kind of vehicle color identification method |
| CN109635825A (en) * | 2018-12-19 | 2019-04-16 | 苏州市科远软件技术开发有限公司 | Vehicle attribute detection method, device and storage medium |
-
2019
- 2019-07-02 CN CN201910590409.6A patent/CN110348505B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140063249A1 (en) * | 2012-09-05 | 2014-03-06 | Motorola Solutions, Inc. | Method, apparatus and system for performing facial recognition |
| CN103996041A (en) * | 2014-05-15 | 2014-08-20 | 武汉睿智视讯科技有限公司 | Vehicle color identification method and system based on matching |
| CN106778583A (en) * | 2016-12-07 | 2017-05-31 | 北京理工大学 | Vehicle attribute recognition methods and device based on convolutional neural networks |
| CN107067011A (en) * | 2017-03-20 | 2017-08-18 | 北京邮电大学 | A kind of vehicle color identification method and device based on deep learning |
| CN108875754A (en) * | 2018-05-07 | 2018-11-23 | 华侨大学 | A kind of vehicle recognition methods again based on more depth characteristic converged network |
| CN109376781A (en) * | 2018-10-24 | 2019-02-22 | 深圳市腾讯网络信息技术有限公司 | A kind of training method, image-recognizing method and the relevant apparatus of image recognition model |
| CN109558872A (en) * | 2018-11-22 | 2019-04-02 | 四川大学 | A kind of vehicle color identification method |
| CN109635825A (en) * | 2018-12-19 | 2019-04-16 | 苏州市科远软件技术开发有限公司 | Vehicle attribute detection method, device and storage medium |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126133A (en) * | 2019-11-08 | 2020-05-08 | 博云视觉（北京）科技有限公司 | Intelligent refrigerator access action recognition method based on deep learning |
| CN111291812A (en) * | 2020-02-11 | 2020-06-16 | 浙江大华技术股份有限公司 | Attribute class acquisition method and device, storage medium and electronic device |
| CN111291812B (en) * | 2020-02-11 | 2023-10-17 | 浙江大华技术股份有限公司 | Method and device for acquiring attribute category, storage medium and electronic device |
| CN112016433A (en) * | 2020-08-24 | 2020-12-01 | 高新兴科技集团股份有限公司 | Vehicle color identification method based on deep neural network |
| CN116563770A (en) * | 2023-07-10 | 2023-08-08 | 四川弘和数智集团有限公司 | Method, device, equipment and medium for detecting vehicle color |
| CN116563770B (en) * | 2023-07-10 | 2023-09-29 | 四川弘和数智集团有限公司 | Method, device, equipment and medium for detecting vehicle color |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110348505B (en) | 2022-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110348505A (en) | Vehicle color disaggregated model training method, device and vehicle color identification method | |
| CN108229362B (en) | Binocular face recognition living body detection method based on access control system | |
| CN108717524B (en) | Gesture recognition system based on double-camera mobile phone and artificial intelligence system | |
| CN112926405B (en) | Method, system, equipment and storage medium for detecting wearing of safety helmet | |
| CN109859171A (en) | A kind of flooring defect automatic testing method based on computer vision and deep learning | |
| CN111444848A (en) | Specific scene model upgrading method and system based on federal learning | |
| CN111860533A (en) | Image recognition method and device, storage medium and electronic device | |
| CN106886778B (en) | License plate character segmentation and recognition method in monitoring scene | |
| CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
| CN108205661A (en) | A kind of ATM abnormal human face detection based on deep learning | |
| CN111401188B (en) | Traffic police gesture recognition method based on human body key point characteristics | |
| CN104951773A (en) | Real-time face recognizing and monitoring system | |
| CN109120919B (en) | A kind of automatic analysis system and method for the evaluation and test of picture quality subjectivity | |
| CN105046219A (en) | Face identification system | |
| CN106339657B (en) | Crop straw burning monitoring method based on monitor video, device | |
| CN104598924A (en) | Target matching detection method | |
| CN107179272B (en) | Acute isolation nerve cell method for catching under a kind of microscope | |
| CN105022999A (en) | Man code company real-time acquisition system | |
| CN106529494A (en) | Human face recognition method based on multi-camera model | |
| CN112837282A (en) | Small sample image defect detection method based on cloud edge cooperation and deep learning | |
| CN105069816B (en) | A kind of method and system of inlet and outlet people flow rate statistical | |
| Amlekar et al. | Leaf features based plant classification using artificial neural network | |
| CN110569780A (en) | high-precision face recognition method based on deep transfer learning | |
| CN113065568A (en) | Target detection, attribute identification and tracking method and system | |
| CN115035088A (en) | Helmet wearing detection method based on yolov5 and posture estimation |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |