CN110348505B - Vehicle color classification model training method and device and vehicle color identification method - Google Patents
Vehicle color classification model training method and device and vehicle color identification method Download PDFInfo
- Publication number
- CN110348505B CN110348505B CN201910590409.6A CN201910590409A CN110348505B CN 110348505 B CN110348505 B CN 110348505B CN 201910590409 A CN201910590409 A CN 201910590409A CN 110348505 B CN110348505 B CN 110348505B
- Authority
- CN
- China
- Prior art keywords
- vehicle color
- vehicle
- color classification
- classification model
- scene
- 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.)
- Active
Links
Images
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 vehicle color classification model training method, which comprises the following steps: inputting the acquired data of a plurality of vehicle images into a pre-trained deep neural network to extract a vehicle color feature set; clustering operation is carried out on the vehicle color feature set by using a clustering algorithm to obtain a plurality of subsets under different scene categories; inputting the vehicle images under the scene categories corresponding to the subsets into corresponding vehicle color classification submodels to train the vehicle color classification submodels; linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model to train the vehicle color classification model; the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network. The invention also discloses a vehicle color classification model training device and a vehicle color identification method. By adopting the embodiment of the invention, the change of the body color in different scenes can be coped with, and the vehicle color identification precision is improved.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a vehicle color classification model training method and device and a vehicle color recognition method.
Background
The vehicle body color identification is a key problem in the field of vehicle attribute identification, is a basis for vehicle feature identification and analysis, and has important research value. With the popularization and the use of deep learning theories, the vehicle detection accuracy is continuously improved, and based on the detection results of the vehicle body or the vehicle face and other parts, the existing vehicle body color identification method mainly focuses on the following modes:
the first mode is the research of color detection area positioning. The precondition for performing the vehicle color recognition is to acquire a stable color recognition area, for example, a partial rectangle on the front cover of the vehicle is used as the color detection area, or the entire vehicle area is used as the color detection area.
And a second mode is based on the research of the whole vehicle image segmentation algorithm. In addition to the method of locating the color recognition area, vehicle segmentation is also the main method of vehicle color recognition. And segmenting the vehicle body part and the background part by using an image processing algorithm so as to identify the color of the vehicle body part.
The vehicle color identification under the complex scene is a research hotspot in the field of vehicle attribute identification, and although certain achievements are achieved, the vehicle color is still difficult to be influenced by scene factors such as illumination, dust and the like, the color is a characteristic with subjectivity, and the condition that human eyes are difficult to identify due to the influence of the scene factors such as illumination, dust and the like on the vehicle color in the monitoring scene often occurs. The color detection area is difficult to accurately position, no matter the color identification area is based on the automobile body or the automobile face under an outdoor complex scene, the influence of a current detection algorithm, the attitude of the automobile, complex background interference or foreground shielding can be applied to the intercepted process, and the intercepted automobile color identification area often has a large amount of information of non-automobile colors.
Disclosure of Invention
The embodiment of the invention aims to provide a vehicle color classification model training method, a vehicle color classification model training device and a vehicle color identification method, which can cope with the change of vehicle body colors in different scenes and improve the vehicle color identification precision.
In order to achieve the above object, an embodiment of the present invention provides a vehicle color classification model training method, including:
inputting a plurality of vehicle images into a pre-trained deep neural network to extract a vehicle color feature set;
clustering operation is carried out on the vehicle color feature set by using a clustering algorithm to obtain a plurality of subsets under different scene categories;
inputting the vehicle images under the scene category corresponding to the subset into a corresponding vehicle color classification submodel to train the vehicle color classification submodel;
linking the output layers of all the vehicle color classification submodels to the input layer of a vehicle color classification model to train the vehicle color classification model; and the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network.
Compared with the prior art, the vehicle color classification model training method disclosed by the invention comprises the steps of firstly, inputting a vehicle image into a pre-trained deep neural network, converting image data into numerical data through the deep neural network, and extracting a vehicle color feature set; then, clustering operation is carried out on the vehicle color feature set by using a clustering algorithm, and the data set can be divided into a plurality of corresponding subsets according to the scene similarity of the images; and finally, respectively training corresponding vehicle color classification submodels according to the subsets, and linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model so as to train the vehicle color classification model. The vehicle color classification model which is trained by integrating the vehicle color classification submodels under all scenes has strong adaptability to illumination of all scenes, can cope with the change of the colors of the vehicle body when the illumination condition changes, and has stronger adaptability and robustness, thereby achieving higher color identification precision.
As an improvement of the above scheme, the clustering algorithm is used to perform clustering operation on the vehicle color feature set to obtain a plurality of subsets under different scene categories, and the formula is satisfied:
wherein, x is set for the vehicle color feature set DiE, D, wherein i is 1 … N, which is the numerical characteristic of one vehicle image; c. CkThe number of the subsets; the set of vehicle color features is divided into a number of the subsets Sj,j∈1,…k。
As an improvement of the scheme, the clustering algorithm is a K-Means algorithm.
As an improvement of the above solution, before the step of linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model, the method further includes:
and deleting the full connection layer and the classification layer in the vehicle color classification submodel.
As an improvement of the scheme, the deep neural network comprises at least one of VGG16, ResNet and SqueezeNet.
In order to achieve the above object, an embodiment of the present invention further provides a vehicle color classification model training device, including:
the vehicle color feature set extraction module is used for inputting a plurality of vehicle images into a pre-trained deep neural network so as to extract a vehicle color feature set;
the clustering operation module is used for carrying out clustering operation on the vehicle color feature set by utilizing a clustering algorithm to obtain a plurality of subsets under different scene categories;
the vehicle color classification submodel training module is used for inputting the vehicle images under the scene categories corresponding to the subsets into the corresponding vehicle color classification submodels so as to train the vehicle color classification submodels;
the vehicle color classification model training module is used for linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model so as to train the vehicle color classification model; and the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network.
Compared with the prior art, the vehicle color classification model training device disclosed by the invention has the advantages that firstly, a vehicle color characteristic set extraction module inputs a vehicle image into a pre-trained deep neural network, image data is converted into numerical data through the neural depth network, and a vehicle color characteristic set is extracted; then, the clustering operation module performs clustering operation on the vehicle color feature set by using a clustering algorithm, and can divide the data set into a plurality of corresponding subsets according to the scene similarity of the images; and finally, the vehicle color classification submodel training module respectively trains corresponding vehicle color classification submodels according to the subsets, and the vehicle color classification model training module links the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model so as to train the vehicle color classification model. The vehicle color classification model which is trained by integrating the vehicle color classification submodels under all scenes has strong adaptability to illumination of all scenes, can cope with the change of the colors of the vehicle body when the illumination condition changes, and has stronger adaptability and robustness, thereby achieving higher color identification precision.
As an improvement of the above scheme, the clustering algorithm is used to perform clustering operation on the vehicle color feature set to obtain a plurality of subsets under different scene categories, and the formula is satisfied:
wherein, x is set for the vehicle color feature set DiE, D, wherein i is 1 … N, which is the numerical characteristic of one vehicle image; c. CkThe number of the subsets; the set of vehicle color features is divided into a number of the subsets Sj,j∈1,…k。
As an improvement of the scheme, the clustering algorithm is a K-Means algorithm.
As an improvement of the above scheme, the apparatus further includes a vehicle color classification submodel processing module, where the vehicle color classification submodel processing module is configured to delete a full connection layer and a classification layer in the vehicle color classification submodel.
In order to achieve the above object, an embodiment of the present invention further provides a vehicle color identification method, including:
acquiring an image to be identified of a vehicle;
inputting the image to be recognized into a vehicle color classification model trained in advance; the method for training the vehicle color classification model is the method for training the vehicle color classification model according to any one of claims 1 to 5;
and acquiring output data of the vehicle color classification model as a result of the vehicle color identification.
Drawings
FIG. 1 is a flowchart of a method for training a color classification model of a vehicle according to an embodiment of the present invention;
FIG. 2 is a training framework of a vehicle color classification submodel in a vehicle color classification model training method according to an embodiment of the present invention;
FIG. 3 is a training framework of a vehicle color classification model in a vehicle color classification model training method according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a vehicle color classification model training apparatus according to an embodiment of the present invention;
fig. 5 is a flowchart of a vehicle color identification method according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It is worth noting that there are often thousands of cameras in a video surveillance system, and the deployment environments of the cameras are not completely consistent, so that there are many differences in image data from the cameras, wherein illumination is the most critical influencing factor. Although the shooting scenes of the images are complex and various, the distribution rule can be still found. In the same time period, due to different illumination conditions, the shot images of the cameras in the same place may have great difference; cameras in different locations may exhibit more consistent image quality due to similar lighting conditions at different time periods. Therefore, in the embodiment of the invention, the corresponding vehicle images are obtained under different illumination scenes for training, so that the vehicle color classification model applicable to various scene categories is trained.
Example one
Referring to fig. 1, fig. 1 is a flowchart of a vehicle color classification model training method according to an embodiment of the present invention; the method comprises the following steps:
s11, inputting a plurality of vehicle images into a pre-trained deep neural network to extract a vehicle color feature set;
s12, carrying out clustering operation on the vehicle color feature set by using a clustering algorithm to obtain a plurality of subsets under different scene categories;
s13, inputting the vehicle images under the scene categories corresponding to the subsets into corresponding vehicle color classification submodels to train the vehicle color classification submodels;
s14, linking the output layers of all the vehicle color classification submodels to the input layer of a vehicle color classification model to train the vehicle color classification model; and the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network.
Specifically, in step S11, the image data cannot be directly clustered, and therefore must be converted into numerical features by means of a deep neural network. Preferably, the deep neural network comprises at least one of VGG16, ResNet, SqueezeNet. Because the color features are the shallow features of the image, the deep neural network can be properly cut in the extraction process. And (3) setting a vehicle color data set as V (namely a plurality of input vehicle images, wherein the vehicle color data set comprises vehicle images of different scene types and different colors), and converting the vehicle images into numerical data by setting a vehicle color feature set extracted by a neural network as D.
Specifically, in step S12, x is set for the set of vehicle color features DiE is equal to D, i is equal to 1 … N and is a numerical characteristic of a vehicle image, and a K-Means algorithm is applied to cluster D so as to cluster images of all similar scenes and meet the formula:
wherein, x is set for the vehicle color feature set DiE, D, wherein i is 1 … N, which is the numerical characteristic of one vehicle image; c. CkThe number of the subsets; the set of vehicle color features is divided into a number of the subsets Sj,j∈1,…k。
The numerical characteristics of the vehicle color images are clustered through K-Means, the data set is divided into a plurality of subsets according to a scene, the intra-class difference of each subset is reduced, performance oscillation is reduced by the deep neural network model trained for each subset, and the classification accuracy can reach a higher level in the scene.
Preferably, the scene types may be divided into a micro-exposure scene, a balanced light scene, an shaded light scene, a weak light scene, and a non-light scene, and in the embodiment of the present invention, the scene types are divided into the above five types, but in other embodiments, the scene types may be set according to actual situations, and are not limited to the above five types of scenes, and all are within the protection scope of the present invention. Through clustering operation, the data set can be divided into a plurality of subsets according to scene similarity of images, images in each subset can come from different cameras or different time periods, but the images have similar scenes and illumination conditions, so that the images have similarity.
Specifically, in step S13, the subset S is targetedjRespectively designing corresponding deep neural networks for training, wherein data input in the training process are vehicle images under the scene types corresponding to the subsets at present, so that k corresponding vehicle color classification submodels are obtained, and the vehicle color classification submodels are uniformly named as a Teacher model. It is worth to be noted that each of the obtained Teacher models can independently identify the vehicle image, for example, when the Teacher model is a vehicle color classification sub-model in a micro-exposure scene, the vehicle image in the micro-exposure scene is input at this time, and color identification can be performed on the vehicle image. The color of the vehicle is identified according to different scenes, so that the problem that the color of the vehicle is difficult to identify due to different illumination can be solved.
As shown in fig. 2, fig. 2 is a training framework of a vehicle color classification submodel in a vehicle color classification model training method according to an embodiment of the present invention; the scene category corresponding to the Teacher1 is a micro-exposure scene, the scene category corresponding to the Teacher2 is a balanced light scene, the scene category corresponding to the Teacher is a shadow light scene, the scene category corresponding to the Teacher4 is a weak light scene, and the scene category corresponding to the Teacher5 is a no light scene. Because the scenes and the illumination of the subsets have certain similarity, each Teacher model can classify the vehicle body color with higher accuracy according to the training subsets of the Teacher model.
Specifically, in step S14, a new deep neural network model, i.e., the vehicle color classification model, is designed based on the obtained k teacher models, and the vehicle color classification model is named Student here. And performing guide training on the Student models by using the k Teacher models, wherein the weight parameters of the Teacher models are not updated any more at this time, only the learned scene knowledge is migrated to the students, and the specific vehicle color classification model framework is shown in fig. 3.
When the student model is trained, the last full connection layer and the softmax classification layer of the 5 teachers models are deleted, and all output layers of the teachers models are linked to the input layer of the student model, so that in the training process, the teachers models can respectively input respective prior knowledge to the student, the feature extraction capability of each teachers model to own scene is transferred to the student model, the student model can adapt to multiple scenes, the change of the color of a vehicle body when the illumination condition changes can be responded, the adaptability and the robustness of the model are stronger, and the higher color recognition accuracy is achieved.
Compared with the prior art, the vehicle color classification model training method disclosed by the invention comprises the steps of firstly, inputting a vehicle image into a pre-trained deep neural network, converting image data into numerical data through the deep neural network, and extracting a vehicle color feature set; then, clustering operation is carried out on the vehicle color feature set by using a clustering algorithm, and the data set can be divided into a plurality of corresponding subsets according to the scene similarity of the images; and finally, respectively training corresponding vehicle color classification submodels according to the subsets, and linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model so as to train the vehicle color classification model. The vehicle color classification model which is trained by integrating the vehicle color classification submodels under all scenes has strong adaptability to illumination of all scenes, can cope with the change of the colors of the vehicle body when the illumination condition changes, and has stronger adaptability and robustness, thereby achieving higher color identification precision.
Example two
Referring to fig. 4, fig. 4 is a schematic structural diagram of a vehicle color classification model training apparatus according to an embodiment of the present invention; the method comprises the following steps:
the vehicle color feature set extraction module 11 is configured to input a plurality of vehicle images into a pre-trained deep neural network to extract a vehicle color feature set;
the clustering operation module 12 is configured to perform clustering operation on the vehicle color feature set by using a clustering algorithm to obtain a plurality of subsets in different scene categories;
a vehicle color classification submodel training module 13, configured to input the vehicle images in the scene category corresponding to the subset into the corresponding vehicle color classification submodel, so as to train the vehicle color classification submodel;
a vehicle color classification model training module 14, configured to link output layers of all the vehicle color classification submodels to an input layer of a vehicle color classification model, so as to train the vehicle color classification model; the vehicle color classification submodel and the vehicle color classification model are both obtained by deep neural network design;
and the vehicle color classification submodel processing module 15 is used for deleting the full connection layer and the classification layer in the vehicle color classification submodel.
Specifically, the image data cannot be directly clustered, and therefore must be converted into numerical features by means of a deep neural network. Preferably, the deep neural network comprises at least one of VGG16, ResNet, SqueezeNet. Because the color features are the shallow features of the image, the deep neural network can be properly cut in the extraction process. The vehicle color feature set extraction module 11 converts the image data into numerical data by taking a vehicle color data set as V (i.e., a plurality of input vehicle images, where the vehicle color data set includes vehicle images of different scene types and different colors), and extracting a vehicle color feature set as D through a neural network.
Specifically, let x be the set of vehicle color features DiE, D, i is 1 … N, the clustering operation module 12 applies K-Means to calculateThe method clusters D, and aims to cluster images of all similar scenes, so that the formula is satisfied:
wherein, x is set for the vehicle color feature set DiE, D, wherein i is 1 … N, which is the numerical characteristic of one vehicle image; c. CkThe number of the subsets; the set of vehicle color features is divided into a number of the subsets Sj,j∈1,…k。
The numerical characteristics of the vehicle color images are clustered through K-Means, the data set is divided into a plurality of subsets according to a scene, the intra-class difference of each subset is reduced, performance oscillation is reduced by the deep neural network model trained for each subset, and the classification accuracy can reach a higher level in the scene.
Preferably, the scene types may be divided into a micro-exposure scene, a balanced light scene, an shaded light scene, a weak light scene, and a non-light scene, and in the embodiment of the present invention, the scene types are divided into the above five types, but in other embodiments, the scene types may be set according to actual situations, and are not limited to the above five types of scenes, and all are within the protection scope of the present invention. Through clustering operation, the data set can be divided into subsets according to scene similarity of images, images in each subset may come from different cameras or different time periods, but the images have similar scenes and illumination conditions, and therefore have similarity.
In particular, the vehicle color classification submodel training module 13 is directed to the subset SjRespectively designing corresponding deep neural networks for training, wherein data input in the training process are vehicle images under the scene types corresponding to the subsets at present, so that k corresponding vehicle color classification submodels are obtained, and the vehicle color classification submodels are uniformly named as a Teacher model.
Specifically, a new deep neural network model, namely the vehicle color classification model, is designed based on the obtained k teacher models, and the vehicle color classification model is named Student here. The vehicle color classification model training module 14 performs guide training on the Student model by using the k Teacher models, and at this time, the weight parameters of the Teacher models are not updated any more, and only the learned scene knowledge is migrated to the Student.
When the student model is trained, the vehicle color classification sub-model processing module 15 deletes the last full connection layer and the softmax classification layer of the 5 teacher models, and the vehicle color classification model training module 14 links the output layers of all the teacher models to the input layer of the student model, so that in the training process, the teacher models can respectively input respective prior knowledge to the students, and the feature extraction capability of each teacher model to the own scene is transferred to the student model, so that the student model can adapt to multiple scenes, can cope with the change of the color of the vehicle body when the illumination condition changes, and has stronger adaptability and robustness, and higher color identification precision is achieved.
Compared with the prior art, the vehicle color classification model training device disclosed by the invention has the advantages that firstly, a vehicle color characteristic set extraction module 11 inputs a vehicle image into a pre-trained deep neural network, image data is converted into numerical data through the neural depth network, and a vehicle color characteristic set is extracted; then, the clustering operation module 12 performs clustering operation on the vehicle color feature set by using a clustering algorithm, and can divide the data set into a plurality of corresponding subsets according to the scene similarity of the images; finally, the vehicle color classification submodel training module 13 trains the corresponding vehicle color classification submodels according to the subsets, and the vehicle color classification model training module 14 links the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model to train the vehicle color classification model. The vehicle color classification model which is trained by integrating the vehicle color classification submodels under all scenes has strong adaptability to illumination of all scenes, can cope with the change of the colors of the vehicle body when the illumination condition changes, and has stronger adaptability and robustness, thereby achieving higher color identification precision.
EXAMPLE III
Referring to fig. 5, fig. 5 is a flowchart of a method for identifying a color of a vehicle according to an embodiment of the present invention; the method comprises the following steps:
s21, acquiring an image to be recognized of the vehicle;
s22, inputting the image to be recognized into a vehicle color classification model trained in advance; the training method of the vehicle color classification model is the training method of the vehicle color classification model according to the first embodiment;
and S23, acquiring the output data of the vehicle color classification model as the result of the vehicle color identification.
The vehicle color recognition method provided by the embodiment of the invention adopts the trained vehicle color classification model training model in the vehicle color classification model training method provided by the embodiment of the invention, so that the change of the body color in different scenes can be coped with, and the vehicle color recognition precision is improved.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (9)
1. A vehicle color classification model training method is characterized by comprising the following steps:
acquiring corresponding vehicle images according to different illumination scenes;
inputting a plurality of vehicle images into a pre-trained deep neural network to extract a vehicle color feature set;
clustering operation is carried out on the vehicle color feature set by using a clustering algorithm to obtain a plurality of subsets under different scene categories; wherein the scene categories include a micro-exposure scene, a balanced light scene, an shaded light scene, a low-light scene, and a no-light scene;
inputting the vehicle images under the scene category corresponding to the subset into a corresponding vehicle color classification submodel to train the vehicle color classification submodel; wherein each vehicle color classification submodel is used for individually identifying a vehicle image;
after deleting the full connection layer and the classification layer in the vehicle color classification submodel, linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model so as to train the vehicle color classification model; and the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network.
2. The vehicle color classification model training method according to claim 1, wherein the clustering algorithm is used for clustering the vehicle color feature set to obtain a plurality of subsets under different scene categories, and the formula is satisfied:
wherein, x is set for the vehicle color feature set DiE, D, wherein i is 1 … N, which is the numerical characteristic of one vehicle image; c. CkThe number of the subsets; the set of vehicle color features is divided into a number of the subsets Sj,j∈1,…k。
3. The vehicle color classification model training method according to claim 2, characterized in that the clustering algorithm is a K-Means algorithm.
4. The vehicle color classification model training method of claim 1, characterized in that the deep neural network comprises at least one of VGG16, ResNet, squeezet.
5. A vehicle color classification model training device, comprising:
the vehicle color feature set extraction module is used for acquiring corresponding vehicle images according to different illumination scenes and inputting a plurality of vehicle images into a pre-trained deep neural network so as to extract a vehicle color feature set;
the clustering operation module is used for carrying out clustering operation on the vehicle color feature set by utilizing a clustering algorithm to obtain a plurality of subsets under different scene categories; wherein the scene categories include a micro-exposure scene, a balanced light scene, an shaded light scene, a low-light scene, and a no-light scene;
the vehicle color classification submodel training module is used for inputting the vehicle images under the scene categories corresponding to the subsets into the corresponding vehicle color classification submodels so as to train the vehicle color classification submodels; wherein each vehicle color classification submodel is used for individually identifying a vehicle image;
the vehicle color classification model training module is used for linking the output layers of all the vehicle color classification submodels to the input layer of the vehicle color classification model after deleting the full connecting layer and the classification layer in the vehicle color classification submodels so as to train the vehicle color classification model; and the vehicle color classification submodel and the vehicle color classification model are obtained by designing a deep neural network.
6. The vehicle color classification model training device according to claim 5, wherein the clustering algorithm is used for clustering the vehicle color feature set to obtain a plurality of subsets under different scene categories, and the formula is satisfied:
7. The vehicle color classification model training device according to claim 6, wherein the clustering algorithm is a K-Means algorithm.
8. The vehicle color classification model training apparatus of claim 6, further comprising a vehicle color classification submodel processing module to delete a fully connected layer and a classification layer in the vehicle color classification submodel.
9. A vehicle color recognition method, characterized by comprising:
acquiring an image to be identified of a vehicle;
inputting the image to be recognized into a vehicle color classification model trained in advance; the method for training the vehicle color classification model is the method for training the vehicle color classification model according to any one of claims 1 to 4;
and acquiring output data of the vehicle color classification model as a result of the vehicle color identification.
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 CN110348505A (en) | 2019-10-18 |
| CN110348505B true 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) |
Families Citing this family (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 |
| 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 |
| 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 |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9471838B2 (en) * | 2012-09-05 | 2016-10-18 | Motorola Solutions, Inc. | Method, apparatus and system for performing facial recognition |
-
2019
- 2019-07-02 CN CN201910590409.6A patent/CN110348505B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110348505A (en) | 2019-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110348505B (en) | Vehicle color classification model training method and device and vehicle color identification method | |
| CN108537136B (en) | Pedestrian re-identification method based on attitude normalization image generation | |
| CN110414368B (en) | Unsupervised pedestrian re-identification method based on knowledge distillation | |
| CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
| CN110458077B (en) | Vehicle color identification method and system | |
| CN108090406B (en) | Face recognition method and system | |
| CN104615986B (en) | The method that pedestrian detection is carried out to the video image of scene changes using multi-detector | |
| Sugiarto et al. | Wood identification based on histogram of oriented gradient (HOG) feature and support vector machine (SVM) classifier | |
| CN112819065B (en) | Unsupervised pedestrian sample mining method and unsupervised pedestrian sample mining system based on multi-clustering information | |
| CN110163236A (en) | The training method and device of model, storage medium, electronic device | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN111027421A (en) | Graph-based direct-push type semi-supervised pedestrian re-identification method | |
| CN111401188B (en) | Traffic police gesture recognition method based on human body key point characteristics | |
| CN110929679A (en) | Non-supervision self-adaptive pedestrian re-identification method based on GAN | |
| CN111046732A (en) | Pedestrian re-identification method based on multi-granularity semantic analysis and storage medium | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| CN113158891B (en) | Cross-camera pedestrian re-identification method based on global feature matching | |
| CN112464730A (en) | Pedestrian re-identification method based on domain-independent foreground feature learning | |
| Anggraini | Face recognition using principal component analysis and self organizing maps | |
| CN109670423A (en) | A kind of image identification system based on deep learning, method and medium | |
| CN113449676B (en) | Pedestrian re-identification method based on two-way interaction-based disentanglement learning | |
| CN110618129A (en) | Automatic power grid wire clamp detection and defect identification method and device | |
| CN112560925A (en) | Complex scene target detection data set construction method and system | |
| CN113743251B (en) | Target searching method and device based on weak supervision scene | |
| Park et al. | Intensity classification background model based on the tracing scheme for deep learning based CCTV pedestrian detection |
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 |