CN106709528A - Method and device of vehicle reidentification based on multiple objective function deep learning - Google Patents
Method and device of vehicle reidentification based on multiple objective function deep learning Download PDFInfo
- Publication number
- CN106709528A CN106709528A CN201710015898.3A CN201710015898A CN106709528A CN 106709528 A CN106709528 A CN 106709528A CN 201710015898 A CN201710015898 A CN 201710015898A CN 106709528 A CN106709528 A CN 106709528A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- deep learning
- learning network
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning 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
Abstract
The present invention discloses a method and device of vehicle reidentification based on multiple objective function deep learning. The method comprises: employing first-class vehicle training data generated based on a vehicle image data set and a preset Softmax objective function to train the preset network parameters in a deep learning network frame, obtaining an optimal depth learning network frame, employing second-class vehicle training data generated based on the vehicle image data and a preset Triplet Loss ternary objective function to train the optimal network parameters in the depth learning network frame, obtaining a target depth learning network frame, and performing reidentification of the vehicle image to be identified according to the target depth learning network frame. Through adoption of the target depth learning network frame, the image features with high robustness and high reliability can be obtained to facilitate vehicle reidentification and effectively satisfy the requirement of the vehicle reidentification in the monitoring video.
Description
Technical field
Recognized again the present invention relates to vehicle identification field, more particularly to a kind of vehicle based on multiple objective function deep learning
Method and device.
Background technology
Vehicular traffic has been used for research vehicle classification, vehicle as the important component in Traffic Surveillance Video
The relevant issues such as detection, vehicle tracking, but recognize that this problem is also a lack of corresponding research again for vehicle.Vehicle is known again
Do not refer in the monitor video gathered from different time and place, to identify same target vehicle.In city scope,
Vehicle is recognized and had great significance for fields such as research intelligent transportation and smart cities again.
Research vehicle recognizes that problem is primarily present following difficult point again：
1) image and video gathered from monitoring system, its is second-rate, so that can make the image of collection has dynamic
The noises such as fuzzy, illumination variation.
2) under monitor video, the similarity between vehicle image is too high.
Due to above-mentioned two difficult point, can only be using the feature of the setting vehicle of the characteristics of image for setting by hand, to carry out
Recognize again, however, being typically histograms of oriented gradients (Histogram of by the characteristics of image figure for setting by hand
Oriented Gradient, HOG) feature, Scale invariant features transform (Scale-invariant feature
Transform, SIFT) feature etc., the vehicle that these characteristics of image cannot be met under monitor video recognizes demand again.
The content of the invention
It is a primary object of the present invention to provide a kind of vehicle based on multiple objective function deep learning again recognition methods and
Device, it is intended to which solution cannot meet the vehicle under monitor video and know again by way of setting characteristics of image by hand in the prior art
The technical problem of other demand.
To achieve the above object, first aspect present invention provides a kind of vehicle based on multiple objective function deep learning and knows again
Other method, the method includes：
Using the first kind vehicle training data based on vehicle image data set generation and preset Softmax object functions
Network parameter in preset deep learning network frame is trained, the deep learning network frame after being optimized；
Using Equations of The Second Kind vehicle training data and preset Triplet based on the vehicle image data set generation
Loss ternarys object function is trained to the network parameter in the deep learning network frame after the optimization, obtains target depth
Degree learning network framework；
Vehicle image to be identified is recognized again according to the target deep learning network framework.
To achieve the above object, second aspect present invention provides a kind of vehicle based on multiple objective function deep learning and knows again
Other device, the device includes：
First training module, for using based on the first kind vehicle training data of vehicle image data set generation and preset
Softmax object functions the network parameter in preset deep learning network frame is trained, the depth after being optimized
Degree learning network framework；
Second training module, for using based on the vehicle image data set generation Equations of The Second Kind vehicle training data and
Preset Triplet Loss ternarys object functions are carried out to the network parameter in the deep learning network frame after the optimization
Training, obtains target deep learning network framework；
Weight identification module, for being known to vehicle image to be identified according to the target deep learning network framework again
Not.
The present invention provides a kind of recognition methods again of the vehicle based on multiple objective function deep learning, and the method includes：Utilize
First kind vehicle training data and preset Softmax object functions based on vehicle image data set generation are to preset depth
Network parameter in learning network framework is trained, the deep learning network frame after being optimized, using based on vehicle figure
As the Equations of The Second Kind vehicle training data and preset Triplet Loss ternarys object functions of data set generation are to the depth after optimization
Network parameter in degree learning network framework is trained, and target deep learning network framework is obtained, according to the target depth
Network frame is practised to recognize vehicle image to be identified again.Relative to prior art, the present invention uses Softmax target letters
Number and Triplet Loss ternarys object functions are trained to deep learning network frame, wherein, using Softmax target letters
It is several that deep learning network frame is trained, the characteristics of image of high robust can be obtained, and in view of due to Softmax mesh
Scalar functions are analysis category differences, and have ignored relation in class, therefore use Triplet Loss ternary object functions
Practise, can effectively increase between class distance and reduce inter- object distance so that can be had using target deep learning network framework
There is the characteristics of image of high robust and high reliability, recognized again in order to carry out vehicle, efficiency and accuracy that raising is recognized again,
Effectively meet the demand that the vehicle under monitor video is recognized again.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those skilled in the art, without having to pay creative labor, can be with root
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is that the flow of the vehicle recognition methods again based on multiple objective function deep learning in first embodiment of the invention is shown
It is intended to；
Fig. 2 is the schematic flow sheet of the refinement step of step 103 in first embodiment of the invention；
Fig. 3 is the function mould of the vehicle weight identifying device based on multiple objective function deep learning in second embodiment of the invention
The schematic diagram of block；
Fig. 4 is the schematic diagram of the refinement functional module of weight identification module 303 in second embodiment of the invention.
Specific embodiment
To enable that goal of the invention of the invention, feature, advantage are more obvious and understandable, below in conjunction with the present invention
Accompanying drawing in embodiment, is clearly and completely described to the technical scheme in the embodiment of the present invention, it is clear that described reality
It is only a part of embodiment of the invention to apply example, and not all embodiments.Based on the embodiment in the present invention, people in the art
The every other embodiment that member is obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
Known again due to the vehicle under monitor video cannot be met by way of setting characteristics of image by hand in the prior art
The technical problem of other demand.
In order to solve the above-mentioned technical problem, the present invention proposes that a kind of vehicle based on multiple objective function deep learning is recognized again
Method, the present invention is using Softmax object functions and Triplet Loss ternarys object function successively to deep learning network frame
Frame is trained, wherein, deep learning network frame is trained using Softmax object functions, robust high can be obtained
Property characteristics of image, and in view of being analysis category difference due to Softmax object functions, and have ignored relation in class, therefore
Learnt using Triplet Loss ternary object functions, can effectively increase between class distance and reduce inter- object distance so that
The characteristics of image with high robust and high reliability can be obtained using the deep learning network frame after second training, so as to
Recognized again in vehicle is carried out, efficiency and accuracy that raising is recognized again.
Fig. 1 is referred to, is that a kind of vehicle based on multiple objective function deep learning is recognized again in first embodiment of the invention
The schematic flow sheet of method, the method includes：
Step 101, using first kind vehicle training data and preset Softmax based on vehicle image data set generation
Object function is trained to the network parameter in preset deep learning network frame, the deep learning network after being optimized
Framework；
Step 102, using the Equations of The Second Kind vehicle training data based on the vehicle image data set generation and preset
Triplet Loss ternarys object functions are trained to the network parameter in the deep learning network frame after optimization, obtain mesh
Mark deep learning network frame；
Step 103, vehicle image to be identified is recognized again according to the target deep learning network framework.
In embodiments of the present invention, deep learning network frame belongs to the one kind in the network structure of neutral net, wherein,
Multiple layer is included in network structure, is sequentially connected between layers, and each layer has input and exports, so as to constitute one
Individual structure end to end.Wherein, the deep learning network frame can be realized using Caffe, or using Torch,
The frameworks such as TensorFlow are realized.
In embodiments of the present invention, with Softmax objective function optimization deep learning network frames, enabling utilize
Deep learning network frame after optimization obtains the characteristics of image of high robust, while, it is contemplated that vehicle recognizes not to be allusion quotation again
The classification problem of type, and Softmax object functions are analysis category differences, and relation in class is have ignored on certain procedures.And
The metric learning of metric form can be very good to solve this problem between study classification.In metric learning, its deep learning
Containing three triples of identical feedforward network in network frame, and instructed using Triplet Loss ternary object functions
Practice, and during being iterated using Triplet Loss ternary object functions, can realize increasing between class distance simultaneously
With the purpose for reducing inter- object distance, even if the present invention is with above-mentioned Softmax object functions and Triplet Loss ternary target letters
It is several that deep learning network frame is trained, to obtain high robust and height using the deep learning network frame after training
The characteristics of image of reliability, is convenient for vehicle and recognizes again.
Due in deep learning network frame, it is necessary to using to two different object functions, accordingly, it would be desirable to pin respectively
Training data is prepared to corresponding object function, wherein, for Softmax object functions, it is necessary to prepare the training of first kind vehicle
Data, include multigroup vehicle image in the first kind vehicle training data, each group of vehicle image N comprising same vehicle is not
With the vehicle image under angle and/or scene, the N is positive integer.It is understood that the mark that the vehicle image of different groups is used
Note is differed, and it is identical to belong to the mark of same group of N vehicle image, for example, the mark of multigroup vehicle image
Respectively label (0), label (1), label (2), label (3) ..., label (i), i is positive integer.For Triplet
Loss ternary object functions, then need to prepare Equations of The Second Kind vehicle training data, and the Equations of The Second Kind vehicle training data includes multigroup three
In tuple vehicle image, and each group of triple vehicle image comprising a target image, with identical with the target image
The positive sample image of mark, and with the negative sample image different with the target image.
Wherein, first kind vehicle training data and Equations of The Second Kind vehicle training data are all based on the car preserved in database
The generation of image data set, i.e. vehicle weight identifying device can be using the vehicle image data set generation first kind vehicle
Training data and Equations of The Second Kind vehicle training data.
Can be pre-stored in database, to be obtained when needing to use.
In embodiments of the present invention, the above-mentioned vehicle based on multiple objective function deep learning again recognition methods by corresponding car
Weight identifying device realize, the vehicle weight identifying device will using based on vehicle image data set generation first kind vehicle train
Data and preset Softmax object functions are trained to the network parameter in preset deep learning network frame, obtain
Deep learning network frame after optimization.Wherein, training method can be the training method of convolutional neural networks.
It is understood that the preset deep learning network frame is ID learning network framework, the depth
There is multilayer in habit network frame, be sequentially connected between layers, and have the network parameter for carrying out computing in each layer.Its
In, the initial value of the network parameter in ID learning network framework is 0.
It should be noted that optimizing instruction to preset deep learning network frame using Softmax object functions
, it is necessary to adjust the number of the neuron of the deep learning network frame when practicing, wherein, number and the first kind vehicle of neuron are instructed
The group number for practicing the vehicle image that packet contains is identical.For example, 100 groups of vehicle images are included in first kind vehicle training data, often
Comprising 8 images under same vehicle difference angle and/or scene in one group of vehicle image, then the number of neuron needs adjustment
It is 100.Additionally, neuron is the junior unit in the Internet of deep learning network frame, neuron represents output vector
Dimension, if for example, in deep learning network frame neuron number be 100, the deep learning network frame export to
The dimension of amount is 100 dimensions.
In embodiments of the present invention, after the deep learning network frame after being optimized, after the optimization is kept
The network structure of deep learning network frame with optimization after network parameter it is constant, recycle be based on vehicle image data set generation
Equations of The Second Kind vehicle training data and preset Triplet Loss ternarys object functions to the deep learning network after the optimization
Network parameter in framework is trained, and obtains target deep learning network framework.
It should be noted that being a net in the framework comprising regularization layer in above-mentioned deep learning network frame
Network layers, are normalized by the network parameter in the deep learning network frame after regularization layer is to optimization so that
Network convergence speed can effectively be accelerated.
Wherein, by using first kind vehicle training data, Softmax object functions to preset deep learning network frame
Frame optimizes training, and using Equations of The Second Kind vehicle training data, Triplet Loss ternarys object functions to optimization training after
Deep learning network frame be trained, obtain target deep learning network framework so that using the target depth learn net
Network framework processes image, can obtain the characteristics of image of high robust and high reliability.
In embodiments of the present invention, after target deep learning network framework is obtained, will be learnt according to the target depth
Network frame is recognized again to vehicle image to be identified.
Further, Fig. 2 is referred to, is the schematic flow sheet of the refinement step of step 103 in first embodiment of the invention,
The refinement step of the step 103 includes：
Step 201, vehicle image to be identified is input into the target deep learning network framework, obtained described to be identified
The characteristics of image of vehicle that includes of vehicle image；
Step 202, recognized again using the characteristics of image of the vehicle.
In embodiments of the present invention, the target depth that vehicle weight identifying device obtains vehicle image input training to be identified
Degree learning network framework, the target deep learning network framework is by the base network parameter that each layer training is obtained in the inner to the vehicle
Image is processed, and exports the characteristics of image that treatment is obtained, so that vehicle weight identifying device gets the characteristics of image, and profit
Recognized again with the characteristics of image.Wherein, identification refers to be carried out with other vehicle image features using the characteristics of image again
Matching, using the vehicle image of Image Feature Matching as same vehicle vehicle image.
In the embodiment of the present invention, vehicle weight identifying device is instructed using the first kind vehicle based on vehicle image data set generation
Practice data and preset Softmax object functions are trained to the network parameter in preset deep learning network frame, obtain
Deep learning network frame after to optimization, using the Equations of The Second Kind vehicle training data based on the vehicle image data set generation
And preset Triplet Loss ternarys object functions are instructed to the network parameter in the deep learning network frame after optimization
Practice, obtain target deep learning network framework, vehicle image to be identified is carried out according to the target deep learning network framework
Recognize again.Relative to prior art, the present invention is using Softmax object functions and Triplet Loss ternary object functions to depth
Degree learning network framework is trained, wherein, deep learning network frame is trained using Softmax object functions, energy
Access the characteristics of image of high robust, and in view of being analysis category difference due to Softmax object functions, and have ignored class
Interior relation, therefore learnt using Triplet Loss ternary object functions, can effectively increase between class distance and reduce class
Interior distance so that can obtain the characteristics of image with high robust and high reliability using target deep learning network framework,
Recognized again in order to carry out vehicle, efficiency and accuracy that raising is recognized again, the vehicle effectively met under monitor video are recognized again
Demand.
Fig. 3 is referred to, is the vehicle weight identifying device based on multiple objective function deep learning in second embodiment of the invention
Functional module schematic diagram, the device includes：
First training module 301, for using based on vehicle image data set generation first kind vehicle training data and
Preset Softmax object functions are trained to the network parameter in preset deep learning network frame, after being optimized
Deep learning network frame；
Second training module 302, for training number using the Equations of The Second Kind vehicle based on the vehicle image data set generation
According to and preset Triplet Loss ternarys object functions to the network parameter in the deep learning network frame after the optimization
It is trained, obtains target deep learning network framework；
Weight identification module 303, for being carried out to vehicle image to be identified according to the target deep learning network framework
Recognize again.
In embodiments of the present invention, deep learning network frame belongs to the one kind in the network structure of neutral net, wherein,
Multiple layer is included in network structure, is sequentially connected between layers, and each layer has input and exports, so as to constitute one
Individual structure end to end.Wherein, the deep learning network frame can be realized using Caffe, or using Torch,
The frameworks such as TensorFlow are realized.
In embodiments of the present invention, with Softmax objective function optimization deep learning network frames, enabling utilize
Deep learning network frame after optimization obtains the characteristics of image of high robust, while, it is contemplated that vehicle recognizes not to be allusion quotation again
The classification problem of type, and Softmax object functions are analysis category differences, and relation in class is have ignored on certain procedures.And
The metric learning of metric form can be very good to solve this problem between study classification.In metric learning, its deep learning
Containing three triples of identical feedforward network in network frame, and instructed using Triplet Loss ternary object functions
Practice, and during being iterated using Triplet Loss ternary object functions, can realize increasing between class distance simultaneously
With the purpose for reducing inter- object distance, even if the present invention is with above-mentioned Softmax object functions and Triplet Loss ternary target letters
It is several that deep learning network frame is trained, to obtain high robust and height using the deep learning network frame after training
The characteristics of image of reliability, is convenient for vehicle and recognizes again.
Due in deep learning network frame, it is necessary to using to two different object functions, accordingly, it would be desirable to pin respectively
Training data is prepared to corresponding object function, wherein, for Softmax object functions, it is necessary to prepare the training of first kind vehicle
Data, include multigroup vehicle image in the first kind vehicle training data, each group of vehicle image N comprising same vehicle is not
With the vehicle image under angle and/or scene, the N is positive integer.It is understood that the mark that the vehicle image of different groups is used
Note is differed, and it is identical to belong to the mark of same group of N vehicle image, for example, the mark of multigroup vehicle image
Respectively label (0), label (1), label (2), label (3) ..., label (i), i is positive integer.For Triplet
Loss ternary object functions, then need to prepare Equations of The Second Kind vehicle training data, and the Equations of The Second Kind vehicle training data includes multigroup three
In tuple vehicle image, and each group of triple vehicle image comprising a target image, with identical with the target image
The positive sample image of mark, and with the negative sample image different with the target image.
In embodiments of the present invention, the first training module 301 is by using the first kind based on vehicle image data set generation
Vehicle training data and preset Softmax object functions are instructed to the network parameter in preset deep learning network frame
Practice, the deep learning network frame after being optimized.Wherein, training method can be the training method of convolutional neural networks.
It is understood that the preset deep learning network frame is ID learning network framework, the depth
There is multilayer in habit network frame, be sequentially connected between layers, and have the network parameter for carrying out computing in each layer.Its
In, the initial value of the network parameter in ID learning network framework is 0.
It should be noted that optimizing instruction to preset deep learning network frame using Softmax object functions
, it is necessary to adjust the number of the neuron of the deep learning network frame when practicing, wherein, number and the first kind vehicle of neuron are instructed
The group number for practicing the vehicle image that packet contains is identical.For example, 100 groups of vehicle images are included in first kind vehicle training data, often
Comprising 8 images under same vehicle difference angle and/or scene in one group of vehicle image, then the number of neuron needs adjustment
It is 100.Additionally, neuron is the junior unit in the Internet of deep learning network frame, neuron represents output vector
Dimension, if for example, in deep learning network frame neuron number be 100, the deep learning network frame export to
The dimension of amount is 100 dimensions.
In embodiments of the present invention, after the deep learning network frame after being optimized, after the optimization is kept
The network structure of deep learning network frame with optimization after network parameter it is constant, the second training module 302 using be based on vehicle
After the Equations of The Second Kind vehicle training data and preset Triplet Loss ternarys object functions of image data set generation are to the optimization
Deep learning network frame in network parameter be trained, obtain target deep learning network framework.
It should be noted that being a net in the framework comprising regularization layer in above-mentioned deep learning network frame
Network layers, are normalized by the network parameter in the deep learning network frame after regularization layer is to optimization so that
Network convergence speed can effectively be accelerated.
Wherein, by using first kind vehicle training data, Softmax object functions to preset deep learning network frame
Frame optimizes training, and using Equations of The Second Kind vehicle training data, Triplet Loss ternarys object functions to optimization training after
Deep learning network frame be trained, obtain target deep learning network framework so that using the target depth learn net
Network framework processes image, can obtain the characteristics of image of high robust and high reliability.
In embodiments of the present invention, after target deep learning network framework is obtained, weight identification module 303 will be according to this
Target deep learning network framework is recognized again to vehicle image to be identified.
Fig. 4 is referred to, is the schematic diagram of the refinement functional module of weight identification module 303 in second embodiment of the invention, should
Weight identification module 303 includes：
Characteristic extracting module 401, for vehicle image to be identified to be input into the target deep learning network framework, obtains
Take the characteristics of image of the vehicle that the vehicle image to be identified is included；
Sub- weight identification module 402, is recognized again for the characteristics of image using the vehicle.
In embodiments of the present invention, the target that characteristic extracting module 401 obtains vehicle image input training to be identified
Deep learning network frame, the target deep learning network framework is by the base network parameter that each layer training is obtained in the inner to the car
Image is processed, and exports the characteristics of image that treatment is obtained, so that characteristic extracting module 401 gets the characteristics of image,
And recognized again using the characteristics of image by sub- weight identification module 402.Wherein, again identification refer to using the characteristics of image and its
His vehicle image feature is matched, using the vehicle image of Image Feature Matching as same vehicle vehicle image.
In the embodiment of the present invention, vehicle weight identifying device is instructed using the first kind vehicle based on vehicle image data set generation
Practice data and preset Softmax object functions are trained to the network parameter in preset deep learning network frame, obtain
Deep learning network frame after to optimization, using the Equations of The Second Kind vehicle training data based on the vehicle image data set generation
And preset Triplet Loss ternarys object functions are instructed to the network parameter in the deep learning network frame after optimization
Practice, obtain target deep learning network framework, vehicle image to be identified is carried out according to the target deep learning network framework
Recognize again.Relative to prior art, the present invention is using Softmax object functions and Triplet Loss ternary object functions to depth
Degree learning network framework is trained, wherein, deep learning network frame is trained using Softmax object functions, energy
Access the characteristics of image of high robust, and in view of being analysis category difference due to Softmax object functions, and have ignored class
Interior relation, therefore learnt using Triplet Loss ternary object functions, can effectively increase between class distance and reduce class
Interior distance so that can obtain the characteristics of image with high robust and high reliability using target deep learning network framework,
Recognized again in order to carry out vehicle, efficiency and accuracy that raising is recognized again, the vehicle effectively met under monitor video are recognized again
Demand.
It is understood that relative to prior art, by the method for ternary metric learning in the embodiment of the present invention
(Triplet Loss ternarys object function) is incorporated into vehicle again in identification field, and combines Softmax object functions, so as to can
To obtain the characteristics of image of high robust and high reliability, apply in vehicle is recognized again, can effectively improve what is recognized again
Accuracy rate, meets the demand that vehicle in video monitoring is recognized again.
In several embodiments provided herein, it should be understood that disclosed apparatus and method, can be by it
Its mode is realized.For example, device embodiment described above is only schematical, for example, the division of the module, only
Only a kind of division of logic function, can there is other dividing mode when actually realizing, such as multiple module or components can be tied
Another system is closed or is desirably integrated into, or some features can be ignored, or do not perform.It is another, it is shown or discussed
Coupling each other or direct-coupling or communication connection can be the INDIRECT COUPLINGs or logical of device or module by some interfaces
Letter connection, can be electrical, mechanical or other forms.
The module that is illustrated as separating component can be or may not be it is physically separate, it is aobvious as module
The part for showing can be or may not be physical module, you can with positioned at a place, or can also be distributed to multiple
On mixed-media network modules mixed-media.Some or all of module therein can be according to the actual needs selected to realize the mesh of this embodiment scheme
's.
In addition, during each functional module in each embodiment of the invention can be integrated in a processing module, it is also possible to
It is that modules are individually physically present, it is also possible to which two or more modules are integrated in a module.Above-mentioned integrated mould
Block can both be realized in the form of hardware, it would however also be possible to employ the form of software function module is realized.
If the integrated module is to realize in the form of software function module and as independent production marketing or use
When, can store in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part for being contributed to prior art in other words or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are used to so that a computer
Equipment (can be personal computer, server, or network equipment etc.) performs the complete of each embodiment methods described of the invention
Portion or part steps.And foregoing storage medium includes：USB flash disk, mobile hard disk, read-only storage (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD etc. are various can store journey
The medium of sequence code.
It should be noted that for foregoing each method embodiment, in order to simplicity is described, therefore it is all expressed as a series of
Combination of actions, but those skilled in the art should know, the present invention not by described by sequence of movement limited because
According to the present invention, some steps can sequentially or simultaneously be carried out using other.Secondly, those skilled in the art should also know
Know, embodiment described in this description belongs to preferred embodiment, and involved action and module might not all be this hairs
Necessary to bright.
In the above-described embodiments, the description to each embodiment all emphasizes particularly on different fields, and does not have the portion described in detail in certain embodiment
Point, may refer to the associated description of other embodiments.
It is more than to a kind of vehicle based on multiple objective function deep learning provided by the present invention again recognition methods and dress
The description put, for those skilled in the art, according to the thought of the embodiment of the present invention, in specific embodiment and range of application
On will change, to sum up, this specification content should not be construed as limiting the invention.
Claims (8)
1. a kind of recognition methods again of the vehicle based on multiple objective function deep learning, it is characterised in that methods described includes：
Using the first kind vehicle training data based on vehicle image data set generation and preset Softmax object functions to pre-
Network parameter in the deep learning network frame put is trained, the deep learning network frame after being optimized；
Using the Equations of The Second Kind vehicle training data based on the vehicle image data set generation and preset TripletLoss ternarys
Object function is trained to the network parameter in the deep learning network frame after the optimization, obtains target depth study net
Network framework；
Vehicle image to be identified is recognized again according to the target deep learning network framework.
2. method according to claim 1, it is characterised in that described to be treated according to the target deep learning network framework
The step of vehicle image of identification is recognized again includes：
Vehicle image to be identified is input into the target deep learning network framework, the vehicle image bag to be identified is obtained
The characteristics of image of the vehicle for containing；
Recognized again using the characteristics of image of the vehicle.
3. method according to claim 1 and 2, it is characterised in that comprising multigroup in the first kind vehicle training data
Vehicle image, N of each group of vehicle image comprising same vehicle opens the vehicle image under different angles and/or scene, and the N is
Positive integer.
4. method according to claim 1 and 2, it is characterised in that multigroup ternary is included in the Equations of The Second Kind training data
Group vehicle image, each group of triple vehicle image comprising a target image, with the target image same tag
Positive sample image, and with the negative sample image different with the target image.
5. a kind of vehicle based on multiple objective function deep learning weighs identifying device, it is characterised in that described device includes：
First training module, for using based on the first kind vehicle training data of vehicle image data set generation and preset
Softmax object functions are trained to the network parameter in preset deep learning network frame, the depth after being optimized
Learning network framework；
Second training module, for using based on the Equations of The Second Kind vehicle training data of the vehicle image data set generation and preset
Triplet Loss ternarys object functions the network parameter in the deep learning network frame after the optimization is trained,
Obtain target deep learning network framework；
Weight identification module, for being recognized to vehicle image to be identified according to the target deep learning network framework again.
6. device according to claim 5, it is characterised in that the heavy identification module includes：
Characteristic extracting module, for vehicle image to be identified to be input into the target deep learning network framework, obtains described
The characteristics of image of the vehicle that vehicle image to be identified is included；
Son weight identification module, is recognized again for the characteristics of image using the vehicle.
7. the device according to claim 5 or 6, it is characterised in that comprising multigroup in the first kind vehicle training data
Vehicle image, N of each group of vehicle image comprising same vehicle opens the vehicle image under different angles and/or scene, and the N is
Positive integer.
8. the device according to claim 5 or 6, it is characterised in that multigroup ternary is included in the Equations of The Second Kind training data
Group vehicle image, each group of triple vehicle image comprising a target image, with the target image same tag
Positive sample image, and with the negative sample image different with the target image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710015898.3A CN106709528A (en) | 2017-01-10 | 2017-01-10 | Method and device of vehicle reidentification based on multiple objective function deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710015898.3A CN106709528A (en) | 2017-01-10 | 2017-01-10 | Method and device of vehicle reidentification based on multiple objective function deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106709528A true CN106709528A (en) | 2017-05-24 |
Family
ID=58907162
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710015898.3A Pending CN106709528A (en) | 2017-01-10 | 2017-01-10 | Method and device of vehicle reidentification based on multiple objective function deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709528A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
| CN108492575A (en) * | 2018-04-11 | 2018-09-04 | 济南浪潮高新科技投资发展有限公司 | A kind of intelligent vehicle type identifier method |
| CN109145777A (en) * | 2018-08-01 | 2019-01-04 | 北京旷视科技有限公司 | Vehicle recognition methods, apparatus and system again |
| CN109165589A (en) * | 2018-08-14 | 2019-01-08 | 北京颂泽科技有限公司 | Vehicle based on deep learning recognition methods and device again |
| CN109948587A (en) * | 2019-03-29 | 2019-06-28 | 清华大学 | Vehicle based on high judgement index feature mining recognition methods and device again |
| CN110619280A (en) * | 2019-08-23 | 2019-12-27 | 长沙千视通智能科技有限公司 | Vehicle heavy identification method and device based on deep joint discrimination learning |
| CN110717402A (en) * | 2019-09-16 | 2020-01-21 | 华中科技大学 | Pedestrian re-identification method based on hierarchical optimization metric learning |
| CN111433843A (en) * | 2017-10-27 | 2020-07-17 | 谷歌有限责任公司 | Unsupervised learning of semantic audio representations |
| CN111967290A (en) * | 2019-05-20 | 2020-11-20 | 阿里巴巴集团控股有限公司 | Object identification method and device and vehicle |
| CN112329569A (en) * | 2020-10-27 | 2021-02-05 | 武汉理工大学 | Freight vehicle state real-time identification method based on image deep learning system |
| US11455807B2 (en) * | 2018-09-20 | 2022-09-27 | Nvidia Corporation | Training neural networks for vehicle re-identification |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030072484A1 (en) * | 2001-09-17 | 2003-04-17 | Kokko Eric Gerard | Method and apparatus for identifying and quantifying characteristics of seeds and other small objects |
| CN104077595A (en) * | 2014-06-15 | 2014-10-01 | 北京工业大学 | Deep belief network image recognition method based on Bayesian regularization |
| CN105184271A (en) * | 2015-09-18 | 2015-12-23 | 苏州派瑞雷尔智能科技有限公司 | Automatic vehicle detection method based on deep learning |
| CN105590102A (en) * | 2015-12-30 | 2016-05-18 | 中通服公众信息产业股份有限公司 | Front car face identification method based on deep learning |
| CN105654066A (en) * | 2016-02-02 | 2016-06-08 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
| CN106096538A (en) * | 2016-06-08 | 2016-11-09 | 中国科学院自动化研究所 | Face identification method based on sequencing neural network model and device |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
-
2017
- 2017-01-10 CN CN201710015898.3A patent/CN106709528A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030072484A1 (en) * | 2001-09-17 | 2003-04-17 | Kokko Eric Gerard | Method and apparatus for identifying and quantifying characteristics of seeds and other small objects |
| CN104077595A (en) * | 2014-06-15 | 2014-10-01 | 北京工业大学 | Deep belief network image recognition method based on Bayesian regularization |
| CN105184271A (en) * | 2015-09-18 | 2015-12-23 | 苏州派瑞雷尔智能科技有限公司 | Automatic vehicle detection method based on deep learning |
| CN105590102A (en) * | 2015-12-30 | 2016-05-18 | 中通服公众信息产业股份有限公司 | Front car face identification method based on deep learning |
| CN105654066A (en) * | 2016-02-02 | 2016-06-08 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
| CN106096538A (en) * | 2016-06-08 | 2016-11-09 | 中国科学院自动化研究所 | Face identification method based on sequencing neural network model and device |
| CN106203533A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | The degree of depth based on combined training study face verification method |
Non-Patent Citations (1)
| Title |
|---|
| 陈雯柏: "《人工神经网络原理与实践》", 31 January 2016, 西安电子科技大学出版社 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111433843A (en) * | 2017-10-27 | 2020-07-17 | 谷歌有限责任公司 | Unsupervised learning of semantic audio representations |
| CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
| CN107886073B (en) * | 2017-11-10 | 2021-07-27 | 重庆邮电大学 | Fine-grained vehicle multi-attribute identification method based on convolutional neural network |
| CN108492575A (en) * | 2018-04-11 | 2018-09-04 | 济南浪潮高新科技投资发展有限公司 | A kind of intelligent vehicle type identifier method |
| CN109145777A (en) * | 2018-08-01 | 2019-01-04 | 北京旷视科技有限公司 | Vehicle recognition methods, apparatus and system again |
| CN109165589A (en) * | 2018-08-14 | 2019-01-08 | 北京颂泽科技有限公司 | Vehicle based on deep learning recognition methods and device again |
| US11455807B2 (en) * | 2018-09-20 | 2022-09-27 | Nvidia Corporation | Training neural networks for vehicle re-identification |
| CN109948587A (en) * | 2019-03-29 | 2019-06-28 | 清华大学 | Vehicle based on high judgement index feature mining recognition methods and device again |
| CN111967290A (en) * | 2019-05-20 | 2020-11-20 | 阿里巴巴集团控股有限公司 | Object identification method and device and vehicle |
| WO2020233414A1 (en) * | 2019-05-20 | 2020-11-26 | 阿里巴巴集团控股有限公司 | Object recognition method and apparatus, and vehicle |
| CN110619280B (en) * | 2019-08-23 | 2022-05-24 | 长沙千视通智能科技有限公司 | Vehicle re-identification method and device based on deep joint discrimination learning |
| CN110619280A (en) * | 2019-08-23 | 2019-12-27 | 长沙千视通智能科技有限公司 | Vehicle heavy identification method and device based on deep joint discrimination learning |
| CN110717402A (en) * | 2019-09-16 | 2020-01-21 | 华中科技大学 | Pedestrian re-identification method based on hierarchical optimization metric learning |
| CN110717402B (en) * | 2019-09-16 | 2022-08-05 | 华中科技大学 | Pedestrian re-identification method based on hierarchical optimization metric learning |
| CN112329569A (en) * | 2020-10-27 | 2021-02-05 | 武汉理工大学 | Freight vehicle state real-time identification method based on image deep learning system |
| CN112329569B (en) * | 2020-10-27 | 2024-02-09 | 武汉理工大学 | Freight vehicle state real-time identification method based on image deep learning system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709528A (en) | Method and device of vehicle reidentification based on multiple objective function deep learning | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN108615010B (en) | Facial expression recognition method based on parallel convolution neural network feature map fusion | |
| CN109685115B (en) | Fine-grained conceptual model with bilinear feature fusion and learning method | |
| CN108491880B (en) | Object classification and pose estimation method based on neural network | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| CN104599275B (en) | The RGB-D scene understanding methods of imparametrization based on probability graph model | |
| Xu et al. | Maize diseases identification method based on multi-scale convolutional global pooling neural network | |
| CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
| CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
| CN106682569A (en) | Fast traffic signboard recognition method based on convolution neural network | |
| CN107145889A (en) | Target identification method based on double CNN networks with RoI ponds | |
| CN109492750B (en) | Zero sample image classification method based on convolutional neural network and factor space | |
| CN110135231A (en) | Animal face recognition methods, device, computer equipment and storage medium | |
| CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
| CN106845528A (en) | A kind of image classification algorithms based on K means Yu deep learning | |
| CN107392241A (en) | A kind of image object sorting technique that sampling XGBoost is arranged based on weighting | |
| CN112614119A (en) | Medical image region-of-interest visualization method, device, storage medium and equipment | |
| CN111126361B (en) | SAR target identification method based on semi-supervised learning and feature constraint | |
| CN110009051A (en) | Feature extraction unit and method, DCNN model, recognition methods and medium | |
| CN104751186A (en) | Iris image quality classification method based on BP (back propagation) network and wavelet transformation | |
| CN105631477A (en) | Traffic sign recognition method based on extreme learning machine and self-adaptive lifting | |
| CN114492634B (en) | Fine granularity equipment picture classification and identification method and system | |
| CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
| CN104573701B (en) | A kind of automatic testing method of Tassel of Corn |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |
|
| RJ01 | Rejection of invention patent application after publication |