CN106529446A - Vehicle type identification method and system based on multi-block deep convolutional neural network - Google Patents
Vehicle type identification method and system based on multi-block deep convolutional neural network Download PDFInfo
- Publication number
- CN106529446A CN106529446A CN201610956426.3A CN201610956426A CN106529446A CN 106529446 A CN106529446 A CN 106529446A CN 201610956426 A CN201610956426 A CN 201610956426A CN 106529446 A CN106529446 A CN 106529446A
- Authority
- CN
- China
- Prior art keywords
- feature
- theta
- convolutional neural
- face image
- neural networks
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a vehicle type identification method and system based on a multi-block deep convolutional neural network. The method comprises: when a target vehicle passes by traffic block port equipment, the location of the target vehicle is localized and a vehicle face image of the target vehicle is intercepted; pretreatment is carried out on the vehicle face image; the target vehicle face image after pretreatment is segmented into a plurality of vehicle face image blocks; a deep convolutional neural network model is established and the plurality of vehicle face image blocks are guided into the deep convolutional neural network model to carry out feature extraction, feature fusion, and feature expansion processing; the expanded vehicle face feature is inputted into a classifier to carry out regression training, thereby realizing vehicle face feature classification identification. According to the method provided by the invention, the target vehicle face image is processed by multi-block processing and the deep convolutional neural network model is established and feature extraction, feature fusion, and feature expansion are carried out on the multi-block target vehicle face images; and then regression training is carried out on the expanded feature by using the classifier, thereby realizing vehicle face feature classification and identification.
Description
Technical field
The invention mainly relates to image identification technical field, and in particular to a kind of to be based on many piecemeal deep layer convolutional neural networks
Model recognizing method and system.
Background technology
The current requirement more and more higher in intelligent transportation system to the robustness and reliability of traffic surveillance and control system, for example:
Fake-licensed car is recognized in traffic system so as to renovate traffic violations even crime runaway convict, car is judged by being identified to vehicle
Whether it is fake-licensed car, this is a kind of effectively method, but using shallow-layer convolution god more than vehicle cab recognition technology traditional at present
Jing networks (CNN), shallow-layer convolutional neural networks (CNN) include an input layer, a convolutional layer, an output layer processing,
The accuracy rate of identification is not high, is unfavorable for judging vehicle.
The content of the invention
The technical problem to be solved is to provide a kind of vehicle based on many piecemeal deep layer convolutional neural networks and knows
Target carriage face image is carried out many piecemeal process, and sets up deep layer convolutional neural networks model from many piecemeals by other method and system
Target carriage face image in extract each feature, and carry out Feature Fusion and characteristic expansion is processed, then by softmax graders
Regression training is carried out to the feature after expansion, the Classification and Identification of car face feature is realized.
The technical scheme that the present invention solves above-mentioned technical problem is as follows:It is a kind of based on many piecemeal deep layer convolutional neural networks
Model recognizing method, comprises the steps:
Step S1:Target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts target vehicle
Car face image, so as to obtain target carriage face image;;
Step S2:Pretreatment is carried out to target carriage face image;
Step S3:Pretreated target carriage face image is divided into into multiple car face image piecemeals;
Step S4:Deep layer convolutional neural networks model is set up, the plurality of car face image piecemeal is imported into the deep layer volume
Feature extraction, Feature Fusion and characteristic expansion are carried out in product neural network model to process, the car face feature after being launched;
Step S5:Regression training will be carried out in car face feature input softmax graders after expansion, realized to the car
Face tagsort is recognized.
The invention has the beneficial effects as follows:Target carriage face image is carried out into many piecemeal process, and sets up deep layer convolutional Neural net
Network model extracts each feature from the target carriage face image of many piecemeals, is obtained in that more texture informations, and carries out feature
Fusion and characteristic expansion are processed, then carry out regression training to the feature after expansion by softmax graders, realize car face feature
Classification and Identification, its accuracy rate be better than conventional sorting methods, the detection and suspect to fake-licensed car with car follow the trail of and search carry
High more correctly judgement rate, plays good income.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, carry out pretreatment to target carriage face picture to concretely comprise the following steps:
Step S301:It is single channel grayscale mode by the RGB patten transformations of target carriage face picture;
Step S302:Target carriage face picture after gradation conversion is carried out by piecemeal according to piecemeal quantity preset value n, so as to
To n block car face image piecemeals.
Using the beneficial effect of above-mentioned further scheme it is:Gray proces are carried out, illumination variation can be reduced and noise is dirty
The impact of dye, improves Stability and veracity the step of extraction characteristics of image after being;Target carriage face picture is carried out into piecemeal,
Process in being easily introduced into deep layer convolutional neural networks model.
Further, in step S301, using formula f (i, j)=0.2999R+0.587G+0.114B by target carriage face
The RGB patten transformations of the target carriage face picture of picture are single channel grayscale mode, wherein, f (i, j) is image coordinate after gray processing
The grey scale pixel value at (i, j) place, R, G, B are respectively coloured image RGB three-components.
Using the beneficial effect of above-mentioned further scheme it is:Target carriage face picture after gray processing can reduce color and light
According to impact, be after extract characteristics of image the step of improve Stability and veracity.
Further, the deep layer convolutional neural networks model includes convolutional neural networks, the concat fusions being sequentially connected
Layer and the full articulamentum of multilamellar, the number of the convolutional neural networks is identical with the number of car face image piecemeal and one-to-one corresponding,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to
Obtain global feature;
The full articulamentum of the multilamellar, the car face feature for the global feature is carried out characteristic expansion, after being launched.
Using the beneficial effect of above-mentioned further scheme it is:Many piecemeal car face image feature extractions, can take into account local special
Levy and global feature, less network parameter can be contained on the basis of feature extraction effect is ensured, efficiency and standard can be reached
Really the optimum of rate, extracts more deep layer texture informations.
Further, regression training will be carried out in the car face feature input softmax graders after expansion in step S5
Specific algorithm be:
The hypothesis function of softmax is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;The parameter of model is represented,Expression is normalized to probability distribution so that all probability
Sum is 1;
When carrying out softmax and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2... θkBy row sieve
Row are obtained:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set
Number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
Using the beneficial effect of above-mentioned further scheme it is:Can autonomic learning feature, by each car face tagsort recognize, obtain
To higher accuracy rate.
Another technical scheme that the present invention solves above-mentioned technical problem is as follows:It is a kind of to be based on many piecemeal deep layer convolutional Neural nets
The model recognition system of network, including:
Image interception module, for target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts
The car face image of target vehicle, so as to obtain target carriage face image;
Image pre-processing module, for carrying out pretreatment to target carriage face image;
Image segmentation module, for pretreated target carriage face image is split, obtains multiple car face images point
Block;
Feature processing block, for setting up deep layer convolutional neural networks model, the plurality of car face image piecemeal is imported
Carry out feature extraction, Feature Fusion and characteristic expansion to process in the deep layer convolutional neural networks model, the car face after expansion is special
Levy;
Classification and Identification module, it is for regression training will be carried out in the car face feature input softmax graders after expansion, real
Now the car face tagsort is recognized.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, described image pretreatment module includes:
Converting unit, for by the RGB patten transformations of target carriage face picture be single channel grayscale mode;
Blocking unit, for the target carriage face picture after gradation conversion is carried out piecemeal according to piecemeal quantity preset value n, from
And obtain n block car face image piecemeals.
Further, in the converting unit, using formula f (i, j)=0.2999R+0.587G+0.114B by target carriage face
The RGB patten transformations of the target carriage face picture of picture are single channel grayscale mode, wherein, f (i, j) is image coordinate after gray processing
The grey scale pixel value at (i, j) place, R, G, B are respectively coloured image RGB three-components.
Further, the deep layer convolutional neural networks model includes convolutional neural networks, the concat fusions being sequentially connected
Layer and the full articulamentum of multilamellar, the number of the convolutional neural networks is identical with the number of car face image piecemeal and one-to-one corresponding,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to
Obtain global feature;
The full articulamentum of the multilamellar, for the global feature to be carried out the car face feature after characteristic expansion is launched.
Further, will carry out back in the car face feature input softmax graders after expansion in the Classification and Identification module
The specific algorithm for returning training is:
Softmax regression models are set up, the hypothesis function of Softmax regression models is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;The parameter of model is represented,Expression probability distribution is normalized so that all probability it
With for 1;
When carrying out Softmax regression models and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2...
θkEnumerate by row and obtain:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set
Number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
Description of the drawings
Fig. 1 provides a kind of side of the model recognizing method based on many piecemeal deep layer convolutional neural networks for the embodiment of the present invention
Method flow chart;
Fig. 2 provides a kind of mould of the model recognition system based on many piecemeal deep layer convolutional neural networks for the embodiment of the present invention
Block block diagram;
Fig. 3 provides the module frame chart of deep layer convolutional neural networks model for the embodiment of the present invention;
Fig. 4 is the data structure diagram of feature extraction and features training in the specific embodiment of the invention.
Specific embodiment
The principle and feature of the present invention are described below in conjunction with accompanying drawing, example is served only for explaining the present invention, and
It is non-for limiting the scope of the present invention.
As shown in figure 1, a kind of model recognizing method based on many piecemeal deep layer convolutional neural networks in the embodiment of the present invention,
Comprise the steps:
Step S1:Target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts target vehicle
Car face image, so as to obtain target carriage face image;
Step S2:Pretreatment is carried out to target carriage face image;
Step S3:Pretreated target carriage face image is divided into into multiple car face image piecemeals;
Step S4:Deep layer convolutional neural networks model is set up, the plurality of car face image piecemeal is imported into the deep layer volume
Feature extraction, Feature Fusion and characteristic expansion are carried out in product neural network model to process, the car face feature after being launched;
Step S5:Regression training will be carried out in car face feature input grader after expansion, realized to the car face feature
Classification and Identification.
Target carriage face image can be carried out many piecemeal process by the embodiment of the present invention, and set up deep layer convolutional neural networks mould
Type carries out feature extraction, Feature Fusion and characteristic expansion from the target carriage face image of many piecemeals and processes, then by grader pair
Feature after expansion carries out regression training, realizes the Classification and Identification of car face feature.
In further embodiment of the present invention, the concrete steps of pretreatment are carried out in step s 2 to target carriage face picture
For:
Step S301:It is single channel grayscale mode by the RGB patten transformations of target carriage face picture;
Step S302:Target carriage face picture after gradation conversion is carried out by piecemeal according to piecemeal quantity preset value n, so as to
To n block car face image piecemeals.
Gray proces are carried out, the impact of illumination variation and sound pollution can be reduced, after being, extract the step of characteristics of image
Suddenly improve Stability and veracity;Target carriage face picture is carried out into piecemeal, is easily introduced in deep layer convolutional neural networks model
Process.
In further embodiment of the present invention, formula f (i, j)=0.2999R+0.587G+ is utilized in step S301
The RGB patten transformations of the target carriage face picture of target carriage face picture are single channel grayscale mode by 0.114B, wherein, f (i, j) is
The grey scale pixel value at image coordinate (i, j) place after gray processing, R, G, B are respectively coloured image RGB three-components.
Target carriage face picture after the algorithm gray processing can reduce the impact of color and illumination, extract image special after being
The step of levying improves Stability and veracity.
In further embodiment of the present invention, the deep layer convolutional neural networks model includes the convolutional Neural net being sequentially connected
Network, concat fused layers and the full articulamentum of multilamellar, the number of the convolutional neural networks are identical with the number of car face image piecemeal
And correspond,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to
Obtain global feature;
The full articulamentum of the multilamellar, the car face feature for the global feature is carried out characteristic expansion, after being launched.
Specifically, as shown in figure 3, the target carriage face picture after gradation conversion is divided into 5 pieces, deep layer convolutional neural networks mould
Type enter to drive a vehicle face feature extraction when, corresponding, convolutional neural networks are set to 5, convolutional neural networks be respectively CNN_1,
CNN_2, CNN_3, CNN_4 and CNN_5,5 convolutional neural networks connect the full articulamentum of multilamellar by concat fused layers, come real
Existing feature extraction, fusion, expansion, then regression training will be carried out in the car face feature output softmax graders after expansion, its energy
Contain less network parameter on the basis of feature extraction effect is ensured, the optimum of efficiency and accuracy rate can be reached.
Above-mentioned many piecemeal car face image feature extracting methods, can take into account local feature and global feature, can ensure special
Contain less network parameter on the basis of levying extraction effect, the optimum of efficiency and accuracy rate can be reached, extract more
Deep layer texture information
In further embodiment of the present invention, by the car face feature input softmax graders after expansion in step S5
In carry out the specific algorithm of regression training and be:
The hypothesis function of softmax is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;The parameter of model is represented,Expression probability distribution is normalized so that all probability it
With for 1;
When carrying out softmax and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2... θkBy row sieve
Row are obtained:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set
Number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
The present invention can autonomic learning feature, by each car face tagsort recognize, obtain higher accuracy rate.
Specifically, as shown in figure 4, wherein, part_2~part_6 is represented the data structure of feature extraction and features training
5 block images of car face, data is the lmdb data forms of each piecemeal, and conv for convolutional layer relu for activation primitive pool is
It is full articulamentum that convolutional layer concat is characterized fused layer fc, and label1 is the label of data, and loss is the loss letter of softmax
Number, for calculating the difference of actual value and predictive value.
As shown in Fig. 2 the embodiment of the present invention provides a kind of vehicle cab recognition system based on many piecemeal deep layer convolutional neural networks
System, including:
Image interception module, for target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts
The car face image of target vehicle, so as to obtain target carriage face image;
Image pre-processing module, for carrying out pretreatment to target carriage face image;
Image segmentation module, for pretreated target carriage face image is split, obtains multiple car face images point
Block;
Feature processing block, for setting up deep layer convolutional neural networks model, the plurality of car face image piecemeal is imported
Carry out feature extraction, Feature Fusion and characteristic expansion to process in the deep layer convolutional neural networks model, the car face after expansion is special
Levy;
Classification and Identification module, it is for regression training will be carried out in the car face feature input softmax graders after expansion, real
Now the car face tagsort is recognized.
Target carriage face image can be carried out many piecemeal process by the embodiment of the present invention, and set up deep layer convolutional neural networks mould
Type carries out feature extraction, Feature Fusion and characteristic expansion from the target carriage face image of many piecemeals and processes, then by grader pair
Feature after expansion carries out regression training, realizes the Classification and Identification of car face feature.
In further embodiment of the present invention, described image pretreatment module includes:
Converting unit, for by the RGB patten transformations of target carriage face picture be single channel grayscale mode;
Blocking unit, for the target carriage face picture after gradation conversion is carried out piecemeal according to piecemeal quantity preset value n, from
And obtain n block car face image piecemeals.
In further embodiment of the present invention, in the converting unit, using formula f (i, j)=0.2999R+0.587G+
The RGB patten transformations of the target carriage face picture of target carriage face picture are single channel grayscale mode by 0.114B, wherein, f (i, j) is
The grey scale pixel value at image coordinate (i, j) place after gray processing, R, G, B are respectively coloured image RGB three-components.
Target carriage face picture effect after gray processing is good, improves accuracy and steady after being the step of extract characteristics of image
It is qualitative.
In further embodiment of the present invention, the deep layer convolutional neural networks model includes the convolutional Neural net being sequentially connected
Network, concat fused layers and the full articulamentum of multilamellar, the number of the convolutional neural networks are identical with the number of car face image piecemeal
And correspond,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to
Obtain global feature;
The full articulamentum of the multilamellar, for the global feature to be carried out the car face feature after characteristic expansion is launched.
Many piecemeal car face image feature extractions, can take into account local feature and global feature, can ensure feature extraction effect
Contain less network parameter on the basis of fruit, the optimum of efficiency and accuracy rate can be reached, more deep layer textures are extracted
Information.
In further embodiment of the present invention, by the car face feature input softmax after expansion in the Classification and Identification module
The specific algorithm that regression training is carried out in grader is:
Softmax regression models are set up, the hypothesis function of Softmax regression models is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;The parameter of model is represented,Expression probability distribution is normalized so that all probability it
With for 1;
When carrying out Softmax regression models and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2...
θkEnumerate by row and obtain:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set
Number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
By set up Softmax regression models can autonomic learning feature, so as to by each car face tagsort recognize, obtain compared with
High accuracy rate.
Specifically, as shown in figure 4, wherein, part_2~part_6 is represented the data structure of feature extraction and features training
5 block images of car face, data is the lmdb data forms of each piecemeal, and conv for convolutional layer relu for activation primitive pool is
It is full articulamentum that convolutional layer concat is characterized fused layer fc, and label1 is the label of data, and loss is the loss letter of softmax
Number, for calculating the difference of actual value and predictive value.
Target carriage face image is carried out many piecemeal process by the present invention, and sets up deep layer convolutional neural networks model from many piecemeals
Target carriage face image in extract each feature, be obtained in that more texture informations, and carry out Feature Fusion and characteristic expansion
Process, then regression training is carried out to the feature after expansion by softmax graders, realize the Classification and Identification of car face feature, its
Accuracy rate is better than conventional sorting methods, and the detection and suspect to fake-licensed car is followed the trail of with car and search for improve and relatively correctly judged
Rate, plays good income.
The foregoing is only presently preferred embodiments of the present invention, not to limit the present invention, all spirit in the present invention and
Within principle, any modification, equivalent substitution and improvements made etc. should be included within the scope of the present invention.
Claims (10)
1. a kind of model recognizing method based on many piecemeal deep layer convolutional neural networks, it is characterised in that comprise the steps:
Step S1:Target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts the car face of target vehicle
Image, so as to obtain target carriage face image;;
Step S2:Pretreatment is carried out to target carriage face image;
Step S3:Pretreated target carriage face image is divided into into multiple car face image piecemeals;
Step S4:Deep layer convolutional neural networks model is set up, the plurality of car face image piecemeal is imported into the deep layer convolution god
Carry out feature extraction, Feature Fusion and characteristic expansion to process in Jing network modeies, the car face feature after being launched;
Step S5:Regression training will be carried out in car face feature input grader after expansion, realized to the car face tagsort
Identification.
2. model recognizing method according to claim 1, it is characterised in that the tool of pretreatment is carried out to target carriage face image
Body step is:
Step S201:It is single channel grayscale mode by the RGB patten transformations of target carriage face picture;
Step S202:Target carriage face picture after gradation conversion is carried out by piecemeal according to piecemeal quantity preset value n, so as to obtain n
Block car face image piecemeal.
3. model recognizing method according to claim 2, it is characterised in that in step S301, using formula f (i,
The RGB patten transformations of target carriage face picture are single channel grayscale mode by j)=0.2999R+0.587G+0.114B, wherein, f
The grey scale pixel value of (i, j) for image coordinate (i, j) place after gray processing, R, G, B are respectively coloured image RGB three-components.
4. model recognizing method according to claim 1, it is characterised in that the deep layer convolutional neural networks model includes
Convolutional neural networks, concat fused layers and the full articulamentum of multilamellar being sequentially connected, the number and car of the convolutional neural networks
The number of face image piecemeal is identical and corresponds,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to obtain
Global feature;
The full articulamentum of the multilamellar, the car face feature for the global feature is carried out characteristic expansion, after being launched.
5. model recognizing method according to claim 1, it is characterised in that in step S5, by the car face after expansion
Regression training is carried out in feature input softmax graders, specific algorithm is:
The hypothesis function of softmax is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...
(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;The parameter of model is represented,Expression probability distribution is normalized so that all probability it
With for 1;
When carrying out softmax and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2... θkEnumerate by row
Arrive:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
6. a kind of model recognition system based on many piecemeal deep layer convolutional neural networks, it is characterised in that include:
Image interception module, for target vehicle position is positioned when target vehicle is through traffic block port equipment, and intercepts target
The car face image of vehicle, so as to obtain target carriage face image;
Image pre-processing module, for carrying out pretreatment to target carriage face image;
Image segmentation module, for pretreated target carriage face image is divided into multiple car face image piecemeals;
Feature processing block, for setting up deep layer convolutional neural networks model, the plurality of car face image piecemeal is imported described
Carry out feature extraction, Feature Fusion and characteristic expansion to process in deep layer convolutional neural networks model, the car face feature after expansion;
Classification and Identification module, for will carry out regression training in the car face feature input grader after expansion, realizes to the car
Face tagsort is recognized.
7. model recognition system according to claim 6, it is characterised in that described image pretreatment module includes:
Converting unit, for by the RGB patten transformations of target carriage face picture be single channel grayscale mode;
Blocking unit, for the target carriage face picture after gradation conversion is carried out piecemeal according to piecemeal quantity preset value n, so as to
To n block car face image piecemeals.
8. model recognition system according to claim 7, it is characterised in that in the converting unit, using formula f (i,
The RGB patten transformations of target carriage face picture are single channel grayscale mode by j)=0.2999R+0.587G+0.114B, wherein, f
The grey scale pixel value of (i, j) for image coordinate (i, j) place after gray processing, R, G, B are respectively coloured image RGB three-components.
9. model recognition system according to claim 6, it is characterised in that the deep layer convolutional neural networks model includes
Convolutional neural networks, concat fused layers and the full articulamentum of multilamellar being sequentially connected, the number and car of the convolutional neural networks
The number of face image piecemeal is identical and corresponds,
The convolutional neural networks, for carrying out feature extraction to car face image piecemeal;
The concat fused layers, for each feature that each convolutional neural networks are extracted is carried out Feature Fusion, so as to obtain
Global feature;
The full articulamentum of the multilamellar, for the global feature to be carried out the car face feature after characteristic expansion is launched.
10. model recognition system according to claim 6, it is characterised in that in the Classification and Identification module, after launching
Car face feature input softmax graders in carry out regression training, specific algorithm is:
Softmax regression models are set up, the hypothesis function of Softmax regression models is
Wherein y represents category, takes k different value, is training set { (x by the car face character representation after expansion(1),y(1)),...
(x(m),y(m)), y(i)∈{1,2,...k};J represents classification, and p (y=j | x) represents the probit of classification j;
The parameter of model is represented,Expression is normalized to probability distribution so that all probability sums are 1;
When carrying out Softmax regression models and returning, by θ with the matrix of k × (n+1) representing, the matrix is by θ1, θ2... θkPress
Row is enumerated and is obtained:
Wherein, θ represents whole model parameters, and T represents transposed matrix, k representation dimensions;
The cost function of softmax is:
Wherein,Weight attenuation term is represented, p (y=j | x) represents the probit of classification j, and m represents training set
Number;
The derivative of J (θ) is:
Finally by J (θ) is minimized, realize recognizing the car face tagsort.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610956426.3A CN106529446A (en) | 2016-10-27 | 2016-10-27 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610956426.3A CN106529446A (en) | 2016-10-27 | 2016-10-27 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106529446A true CN106529446A (en) | 2017-03-22 |
Family
ID=58326882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610956426.3A Pending CN106529446A (en) | 2016-10-27 | 2016-10-27 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106529446A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106991420A (en) * | 2017-03-27 | 2017-07-28 | 新智认知数据服务有限公司 | A kind of detection method of license plate of the license plate area regression technique based on piecemeal |
CN107085074A (en) * | 2017-04-19 | 2017-08-22 | 中国科学技术大学 | A kind of method for monitoring motor-vehicle tail-gas of classifying |
CN107239514A (en) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | A kind of plants identification method and system based on convolutional neural networks |
CN107437110A (en) * | 2017-07-11 | 2017-12-05 | 中国科学院自动化研究所 | The piecemeal convolution optimization method and device of convolutional neural networks |
CN107665353A (en) * | 2017-09-15 | 2018-02-06 | 平安科技(深圳)有限公司 | Model recognizing method, device, equipment and computer-readable recording medium based on convolutional neural networks |
CN108319977A (en) * | 2018-01-30 | 2018-07-24 | 浙江大学 | Cervical biopsy area recognizing method based on the multi-modal network of channel information and device |
CN108334892A (en) * | 2017-12-26 | 2018-07-27 | 新智数字科技有限公司 | A kind of model recognizing method, device and equipment based on convolutional neural networks |
CN108681707A (en) * | 2018-05-15 | 2018-10-19 | 桂林电子科技大学 | Wide-angle model recognizing method and system based on global and local Fusion Features |
CN109100648A (en) * | 2018-05-16 | 2018-12-28 | 上海海事大学 | Ocean current generator impeller based on CNN-ARMA-Softmax winds failure fusion diagnosis method |
CN109308678A (en) * | 2017-07-28 | 2019-02-05 | 株式会社理光 | The method, device and equipment relocated using panoramic picture |
CN109685058A (en) * | 2017-10-18 | 2019-04-26 | 杭州海康威视数字技术股份有限公司 | A kind of images steganalysis method, apparatus and computer equipment |
CN109948610A (en) * | 2019-03-14 | 2019-06-28 | 西南交通大学 | A kind of vehicle fine grit classification method in the video based on deep learning |
CN110175622A (en) * | 2019-04-04 | 2019-08-27 | 华中科技大学 | The vehicle part recognition methods of convolutional neural networks based on symbiosis and system |
CN110555877A (en) * | 2018-05-31 | 2019-12-10 | 杭州海康威视数字技术股份有限公司 | Image processing method, device and equipment and readable medium |
CN114943735A (en) * | 2022-07-22 | 2022-08-26 | 浙江省肿瘤医院 | Tongue picture image-based tumor prediction system and method and application thereof |
CN115761659A (en) * | 2023-01-09 | 2023-03-07 | 南京隼眼电子科技有限公司 | Recognition model construction method, vehicle type recognition method, electronic device, and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN105938560A (en) * | 2016-03-23 | 2016-09-14 | 吉林大学 | Convolutional-neural-network-based vehicle model refined classification system |
CN105956560A (en) * | 2016-05-06 | 2016-09-21 | 电子科技大学 | Vehicle model identification method based on pooling multi-scale depth convolution characteristics |
CN105975915A (en) * | 2016-04-28 | 2016-09-28 | 大连理工大学 | Front vehicle parameter identification method based on multitask convolution nerve network |
-
2016
- 2016-10-27 CN CN201610956426.3A patent/CN106529446A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN105938560A (en) * | 2016-03-23 | 2016-09-14 | 吉林大学 | Convolutional-neural-network-based vehicle model refined classification system |
CN105975915A (en) * | 2016-04-28 | 2016-09-28 | 大连理工大学 | Front vehicle parameter identification method based on multitask convolution nerve network |
CN105956560A (en) * | 2016-05-06 | 2016-09-21 | 电子科技大学 | Vehicle model identification method based on pooling multi-scale depth convolution characteristics |
Non-Patent Citations (1)
Title |
---|
张文达 等: "基于多尺度分块卷积神经网络的图像目标识别算法", 《计算机应用》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106991420A (en) * | 2017-03-27 | 2017-07-28 | 新智认知数据服务有限公司 | A kind of detection method of license plate of the license plate area regression technique based on piecemeal |
CN107085074B (en) * | 2017-04-19 | 2019-07-23 | 中国科学技术大学 | A method of classification monitoring motor-vehicle tail-gas |
CN107085074A (en) * | 2017-04-19 | 2017-08-22 | 中国科学技术大学 | A kind of method for monitoring motor-vehicle tail-gas of classifying |
CN107239514A (en) * | 2017-05-19 | 2017-10-10 | 邓昌顺 | A kind of plants identification method and system based on convolutional neural networks |
CN107437110B (en) * | 2017-07-11 | 2021-04-02 | 中国科学院自动化研究所 | Block convolution optimization method and device of convolutional neural network |
CN107437110A (en) * | 2017-07-11 | 2017-12-05 | 中国科学院自动化研究所 | The piecemeal convolution optimization method and device of convolutional neural networks |
CN109308678A (en) * | 2017-07-28 | 2019-02-05 | 株式会社理光 | The method, device and equipment relocated using panoramic picture |
CN109308678B (en) * | 2017-07-28 | 2023-10-27 | 株式会社理光 | Method, device and equipment for repositioning by using panoramic image |
CN107665353A (en) * | 2017-09-15 | 2018-02-06 | 平安科技(深圳)有限公司 | Model recognizing method, device, equipment and computer-readable recording medium based on convolutional neural networks |
US11347977B2 (en) | 2017-10-18 | 2022-05-31 | Hangzhou Hikvision Digital Technology Co., Ltd. | Lateral and longitudinal feature based image object recognition method, computer device, and non-transitory computer readable storage medium |
CN109685058A (en) * | 2017-10-18 | 2019-04-26 | 杭州海康威视数字技术股份有限公司 | A kind of images steganalysis method, apparatus and computer equipment |
CN109685058B (en) * | 2017-10-18 | 2021-07-09 | 杭州海康威视数字技术股份有限公司 | Image target identification method and device and computer equipment |
CN108334892A (en) * | 2017-12-26 | 2018-07-27 | 新智数字科技有限公司 | A kind of model recognizing method, device and equipment based on convolutional neural networks |
CN108334892B (en) * | 2017-12-26 | 2020-11-17 | 新智数字科技有限公司 | Vehicle type identification method, device and equipment based on convolutional neural network |
CN108319977B (en) * | 2018-01-30 | 2020-11-10 | 浙江大学 | Cervical biopsy region identification method and device based on channel information multi-mode network |
CN108319977A (en) * | 2018-01-30 | 2018-07-24 | 浙江大学 | Cervical biopsy area recognizing method based on the multi-modal network of channel information and device |
CN108681707A (en) * | 2018-05-15 | 2018-10-19 | 桂林电子科技大学 | Wide-angle model recognizing method and system based on global and local Fusion Features |
CN109100648B (en) * | 2018-05-16 | 2020-07-24 | 上海海事大学 | CNN-ARMA-Softmax-based ocean current generator impeller winding fault fusion diagnosis method |
CN109100648A (en) * | 2018-05-16 | 2018-12-28 | 上海海事大学 | Ocean current generator impeller based on CNN-ARMA-Softmax winds failure fusion diagnosis method |
CN110555877A (en) * | 2018-05-31 | 2019-12-10 | 杭州海康威视数字技术股份有限公司 | Image processing method, device and equipment and readable medium |
CN109948610A (en) * | 2019-03-14 | 2019-06-28 | 西南交通大学 | A kind of vehicle fine grit classification method in the video based on deep learning |
CN110175622A (en) * | 2019-04-04 | 2019-08-27 | 华中科技大学 | The vehicle part recognition methods of convolutional neural networks based on symbiosis and system |
CN114943735A (en) * | 2022-07-22 | 2022-08-26 | 浙江省肿瘤医院 | Tongue picture image-based tumor prediction system and method and application thereof |
CN115761659A (en) * | 2023-01-09 | 2023-03-07 | 南京隼眼电子科技有限公司 | Recognition model construction method, vehicle type recognition method, electronic device, and storage medium |
CN115761659B (en) * | 2023-01-09 | 2023-04-11 | 南京隼眼电子科技有限公司 | Recognition model construction method, vehicle type recognition method, electronic device, and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106529446A (en) | Vehicle type identification method and system based on multi-block deep convolutional neural network | |
CN104050471B (en) | Natural scene character detection method and system | |
CN105956626A (en) | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method | |
CN107194398A (en) | Car damages recognition methods and the system at position | |
CN111008633B (en) | License plate character segmentation method based on attention mechanism | |
WO2021103897A1 (en) | License plate number recognition method and device, electronic device and storage medium | |
CN104915643A (en) | Deep-learning-based pedestrian re-identification method | |
CN109255284A (en) | A kind of Activity recognition method of the 3D convolutional neural networks based on motion profile | |
CN108009548A (en) | A kind of Intelligent road sign recognition methods and system | |
CN110163069A (en) | Method for detecting lane lines for assisting driving | |
CN112487862A (en) | Garage pedestrian detection method based on improved EfficientDet model | |
CN111639524A (en) | Automatic driving image semantic segmentation optimization method | |
CN115049534A (en) | Knowledge distillation-based real-time semantic segmentation method for fisheye image | |
CN113076994A (en) | Open-set domain self-adaptive image classification method and system | |
CN114418895A (en) | Driving assistance method and device, vehicle-mounted device and storage medium | |
CN114677687A (en) | ViT and convolutional neural network fused writing brush font type rapid identification method | |
CN115019279A (en) | Context feature fusion method based on MobileNet lightweight network | |
CN111462140A (en) | Real-time image instance segmentation method based on block splicing | |
CN115620180A (en) | Aerial image target detection method based on improved YOLOv5 | |
CN111091131A (en) | Self-adaptive license plate character recognition system and recognition method based on multi-task learning | |
CN112052829B (en) | Pilot behavior monitoring method based on deep learning | |
CN111612803A (en) | Vehicle image semantic segmentation method based on image definition | |
CN116682090A (en) | Vehicle target detection method based on improved YOLOv3 algorithm | |
CN112750128A (en) | Image semantic segmentation method and device, terminal and readable storage medium | |
CN116363072A (en) | Light aerial image detection method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |