CN109359696A - A kind of vehicle money recognition methods, system and storage medium - Google Patents
A kind of vehicle money recognition methods, system and storage medium Download PDFInfo
- Publication number
- CN109359696A CN109359696A CN201811268068.2A CN201811268068A CN109359696A CN 109359696 A CN109359696 A CN 109359696A CN 201811268068 A CN201811268068 A CN 201811268068A CN 109359696 A CN109359696 A CN 109359696A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- layer
- vehicle image
- classification
- granularity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to field of image processing, a kind of vehicle money recognition methods, system and storage medium are disclosed, this method comprises: extracting the global characteristics and multiple local features of vehicle image using more granularity neural networks；The global characteristics of the vehicle image and multiple local features are merged using Fusion Features neural network, obtain the corresponding category feature of the vehicle image；According to the category feature, the probability that the vehicle image belongs to each vehicle money classification is obtained using Classification Neural；The highest vehicle money of probability is determined as the corresponding vehicle money of the vehicle image.Vehicle money recognition methods in the present invention is extracted feature using more granularity neural networks of global and local mixing, and is merged to global characteristics and local feature, and the vehicle money classification of vehicle image can be recognized accurately.
Description
Technical field
The present invention relates to field of image processings, in particular to a kind of vehicle money recognition methods, system and storage medium.
Background technique
With the rise of artificial intelligence trend, the application of car tracing and vehicle retrieval technology in real life is increasingly
Extensively, have become urban traffic safety, urban transportation is administered and the illegal important means pursued and captured an escaped prisoner.Theoretically license plate number conduct
The unique identities mark of vehicle, can be solved perfectly car tracing and vehicle retrieval problem, however in real life usually because
To block, illumination, motion blur and the artificial reasons such as forge can not obtain complete and correct license plate number.Vehicle style conduct
Another important feature of vehicle can avoid the influence of above-mentioned unfavorable factor well, can be in conjunction with attributes such as vehicle colors
Car tracing and vehicle retrieval are realized in certain time and spatial dimension, therefore vehicle money identification technology is more and more important, becomes vehicle
Tracking and the essential module of vehicle retrieval system.
Summary of the invention
The purpose of the present invention is to provide a kind of vehicle money recognition methods, system and storage mediums, to solve the above problems.
In a first aspect, the present invention provides a kind of vehicle money recognition methods, comprising:
Using more granularity neural networks, the global characteristics and multiple local features of vehicle image are extracted；
The global characteristics of the vehicle image and multiple local features are carried out using Fusion Features neural network
Fusion, obtains the corresponding category feature of the vehicle image；
According to the category feature, the vehicle image is obtained using Classification Neural and belongs to the general of each vehicle money classification
Rate；
The highest vehicle money of probability is determined as the corresponding vehicle money of the vehicle image.
Optionally, more granularity neural networks are mentioned including the first convolutional layer, maximum pond layer, at least one foundation characteristic
Modulus block and at least three space characteristics extraction modules, first convolutional layer, the maximum pond layer, it is described at least one
Foundation characteristic extraction module is sequentially connected, and the output of at least one foundation characteristic extraction module is respectively with described at least three
The connection of space characteristics extraction module；The foundation characteristic extraction module is used to extract the foundation characteristic in vehicle image, Mei Gesuo
Space characteristics extraction module is stated for learning corresponding global characteristics or local feature, the foundation characteristic extraction module includes
Conv_block and at least one identity_block, the conv_block and at least one described identity_block
It is sequentially connected.
Optionally, each space characteristics extraction module includes average pond layer, full articulamentum and softmax activation
Layer, the average pond layer, the full articulamentum and the softmax active coating are sequentially connected.
Optionally, the Fusion Features neural network includes splicing layer, dimension transform layer, the second convolutional layer, third convolution
Layer, Volume Four lamination and the 5th convolutional layer, the splicing layer, the dimension transform layer, second convolutional layer, third volume
Lamination, the Volume Four lamination and the 5th convolutional layer are sequentially connected, and the splicing layer is used for more granularity nerve nets
Multiple outputs of network are spliced, and the dimension transform layer is used to the output of the splicing layer carrying out dimension transformation.
Optionally, the convolution kernel of second convolutional layer is identical as the convolution kernel of the Volume Four lamination, the third volume
The convolution kernel of lamination is identical as the convolution kernel of the 5th convolutional layer, the convolution kernel of second convolutional layer, the third convolution
Convolution kernel, the convolution kernel of the Volume Four lamination and the convolution kernel of the 5th convolutional layer of layer are asymmetric convolution kernel.
Optionally, described according to the category feature, the vehicle image, which is obtained, using Classification Neural belongs to each vehicle
The probability of money classification, specifically: the Classification Neural merges each space point in the category feature by the 6th convolutional layer
The feature of branch, exports the probability that the vehicle image belongs to each vehicle money classification by softmax active coating.
Optionally, utilize more granularity neural networks, extract vehicle image global characteristics and multiple local features it
Before, the method also includes: more granularity neural networks are trained using the training sample in training set until meeting
Training termination condition.
Optionally, more granularity neural networks are trained in the training sample using in training set until full
After foot training termination condition, the method also includes: the parameter in fixed more granularity neural networks utilizes the training
Sample is trained the Fusion Features neural network and the Classification Neural.
Second aspect, the present invention provide a kind of vehicle money identifying system, comprising:
Characteristic extracting module, for by more granularity neural networks, extract vehicle image global characteristics and multiple offices
Portion's feature；Fusion Features module, for by Fusion Features neural network to the global characteristics of the vehicle image and
Multiple local features are merged, and the corresponding category feature of the vehicle image is obtained；Categorization module, for according to the classification
Feature obtains the probability that the vehicle image belongs to each vehicle money classification by Classification Neural；Vehicle money determining module, by probability
Highest vehicle money is determined as the corresponding vehicle money of the vehicle image.
The third aspect, the present invention provide a kind of electronic equipment, comprising: processor and memory, the memory are stored with
The executable machine readable instructions of processor, execute side described in first aspect when the machine readable instructions are executed by processor
Method.
Fourth aspect, the present invention provide a kind of storage medium, and computer program, the meter are stored on the storage medium
Method described in first aspect is executed when calculation machine program is run by processor.
Compared with the prior art, vehicle money recognition methods provided by the invention passes through sequentially connected more granularity neural networks, spy
Levying fused neural network, Classification Neural can be other to the vehicle image of input progress vehicle inscriptions, passes through more granularity neural networks
The global generic features and local feature for extracting image carry out global characteristics and local feature by Fusion Features neural network
Fusion, and determine that vehicle in picture belongs to the probability of each vehicle money classification by Classification Neural, whole process recognition speed is fast,
And do not influenced by vehicle color and shooting visual angle, for partial occlusion, the factors such as motion blur and light dark influence, equal energy
Reach higher recognition correct rate.
To enable the above objects, features and advantages of the present invention to be clearer and more comprehensible, alternative embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to use required in embodiment in order to illustrate more clearly of the technical solution of embodiment of the present invention
Attached drawing be briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not to be seen as
It is the restriction to range, it for those of ordinary skill in the art, without creative efforts, can be with root
Other relevant attached drawings are obtained according to these attached drawings.
Fig. 1 shows the flow chart of vehicle money recognition methods provided by the present invention；
Fig. 2 shows the structure charts of more granularity neural networks provided by the present invention；
Fig. 3 shows the detailed structure view of more granularity neural networks provided by the present invention；
Fig. 4 shows the structure chart of conv_block and identity_block；
Fig. 5 shows the structure chart of Fusion Features neural network provided by the present invention；
Fig. 6 shows the detailed structure view of Fusion Features neural network and sorter network provided by the present invention；
Fig. 7 shows another flow chart of vehicle money recognition methods provided by the present invention；
Fig. 8 shows the schematic diagram of vehicle money identifying system provided by the present invention.
Icon:
First convolutional layer -10；Maximum pond layer -11；Foundation characteristic extraction module -12；Space characteristics extraction module -13；
Splice layer -20；Dimension transform layer -21；Second convolutional layer -22；Third convolutional layer -23；Volume Four lamination -24；5th convolutional layer-
25；Characteristic extracting module -30；Fusion Features module -31；Categorization module -32；Vehicle money determining module -33.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed invention to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
Range, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
With reference to the accompanying drawing, it elaborates to some embodiments of the present invention.In the absence of conflict, following
Feature in embodiment and embodiment can be combined with each other.
First embodiment
Artificial neural network is the research hotspot of artificial intelligence field at this stage, from information processing angle to human brain neuron
Network is abstracted, and by establishing certain naive model, different neural networks is formed according to different connection types, currently,
Depth learning technology based on artificial neural network achieves progress in every field, including image classification, recognition of face,
Speech recognition etc., the advantage of deep learning are that can use general structure solves different technologies with relatively similar system
Problem.The present embodiment provides a kind of vehicle money recognition methods by multiple neural networks, main in existing vehicle money identification technology
Have following several.
The first, is identified based on generic object identification model, other in vehicle inscriptions using general object identification frame
Training obtains final model on data set, and this method progress vehicle money recognition correct rate is lower, because of general object identification class
Differ greatly between not, and vehicle inscriptions it is different classes of between difference it is smaller, and require with a vehicle in different colours and not
With same class is identified as under shooting visual angle, general object identification frame can not well solve this problem at present.
Second, it is based on component home identification model, by detecting the components such as car light, vehicle window, wheel, headstock and the tailstock,
It extracts the feature of each component home and is merged, then carry out Classification and Identification, however the view shot using classifier
Angle influences and can not detect all specified component homes, can this method be failed, while this method is needed in application
Feature identification is extracted after first detecting each component, identification process is complicated and recognition time is long, it is more difficult to realize real-time car tracing.
The third, is identified based on 3D Model Matching, by estimating the 3D pose of vehicle, under close pose in 3D model library
Perspective view carry out match cognization.This method can be well solved the problem of shooting visual angle, however can not cope with illumination and movement
The problems such as fuzzy, therefore recognition correct rate is lower in reality scene.
The present embodiment is using the completion of multiple neural networks to the extraction of the feature of the vehicle image of input, fusion and vehicle money
Identification, referring to Fig. 1, this method comprises:
Step 101: utilizing more granularity neural networks, extract the global characteristics and multiple local features of vehicle image.
Step 102: the global characteristics of vehicle image and multiple local features being carried out using Fusion Features neural network
Fusion obtains the corresponding category feature of vehicle image.
Step 103: according to the corresponding category feature of vehicle image, obtaining vehicle image using Classification Neural and belong to respectively
The probability of vehicle money classification.
Step 104: the highest vehicle money of probability is determined as the corresponding vehicle money of vehicle image.
More granularity neural networks, Fusion Features neural network, Classification Neural are sequentially connected, wherein the first order it is more
Granularity neural network is used to carry out the abstract of different spaces position to the vehicle image of input, and the overall situation for extracting vehicle image is special
It seeks peace local feature, global characteristics are generic features on the whole, and local feature is the minutia of each area of space, will be global
It is combined together with the feature of part, can there is richer information and details to remove characterization input picture, if merely by straight
The global characteristics for connecing extraction vehicle are carried out classification output, due to being easy to neglect each partial zones in image in global characteristics
The detailed information in domain, therefore often final recognition effect is not satisfactory, the present embodiment is used global and local mixed mixing
The more granularity neural networks of type can be concerned about the global Global Information and local detailed information of image, so that more granularities simultaneously
Neural network can learn into image more representative information.
Vehicle image is inputted in more granularity neural networks, before which can be in a traffic control system
The vehicle region part that the vehicle detection model of grade detects, is also possible to the vehicle pictures of reality scene, passes through more
It spends neural network and foundation characteristic extraction is carried out to the vehicle image of input, and global characteristics to image and local feature be not by
With branch learnt respectively, only need to pay close attention to corresponding feature in each branch and learnt, the output of global branch and more
The output of a localized branches is connected with the input of Fusion Features neural network, by Fusion Features neural network to vehicle image
Each characteristic pattern merged, fused category feature is exported into Classification Neural, and by classification nerve net
Vehicle belongs to the probability of each vehicle money classification in network output image.
Specifically, referring to Fig.2, above-mentioned more granularity neural networks include the first convolutional layer 10, maximum pond layer 11, at least one
A foundation characteristic extraction module 12 and at least three space characteristics extraction modules 13, the first convolutional layer 10, maximum pond layer 11,
At least one foundation characteristic extraction module 12 is sequentially connected, the output of foundation characteristic extraction module 12 respectively at least three spaces
Each of characteristic extracting module 13 is attached, wherein foundation characteristic extraction module 12 is used to extract the vehicle figure of input
As in foundation characteristic, profile, edge and texture including each part of vehicle position etc., the characteristic pattern output it in the horizontal direction into
Row is split, and obtains multiple space position branches, extracts corresponding space position respectively by different space characteristics extraction modules 13
Feature.
It should be noted that only one foundation characteristic may be implemented when the input smaller background of vehicle image size is simple
Extraction module 12 extracts the foundation characteristic of the vehicle image, and still, in practical application, the size of the vehicle image of input is universal
Larger, background is complex, and therefore, to obtain better foundation characteristic extraction effect, adaptability increases foundation characteristic and extracts
The quantity of module 12.
Each foundation characteristic extraction module 12 includes conv_block and at least one identity_block,
In, conv_block and identity_block are sequentially connected, this kind of structure is identical as ResNet50 in residual error network.In residual error
It include two basic block in network, one is conv_block, for reducing the size of characteristic pattern, is removed in characteristic pattern
Lengthy and jumbled information extracts more crucial characteristic information, and the dimension output and input is different, so cannot continuously go here and there
Connection, the other is identity_block, for from complex background, study to be arrived more representative in the similar position of different vehicle
Characteristic information, the characteristic pattern size output and input is constant.When input picture background it is complex, or work as multiple classifications
Vehicle it is more similar when, in order to extract more representational characteristic information, can connect after conv_block multiple
identity_block。
Fig. 3 is a kind of specific connection structure of more granularity neural networks provided in this embodiment, using four foundation characteristics
Extraction module, Block2, Block3, Block4 and Block5 in respectively Fig. 3 are sequentially connected between each Block, pass through
Multiple Block extract the feature in vehicle image, obtain the foundation characteristic of image and are exported, are wrapped in Block2
Include conv_block and two group of identity_block being sequentially connected in series, include in Block3 the conv_block being sequentially connected in series and
It include conv_block and three group of identity_block being sequentially connected in series in three groups of identity_block, Block4,
Include conv_block and two group of identity_block being sequentially connected in series in Block5, the vehicle image is exported in Block5
Foundation characteristic.
Specifically, each conv_ in Block2, Block3, Block4 and Block5 is shown in FIG. 4 in the present embodiment
The structure of block and identity_block includes sequentially connected three convolutional layers, each convolution in conv_block
Convolution kernel size is respectively 1 × 1,3 × 3 and 1 × 1 in layer, and by the output of the convolutional layer of the last layer and another volume 1 × 1
The output of product network is input to relu activation primitive layer, the input and the input of conv_block of 1 × 1 convolutional network after being added
Connection；It include sequentially connected three convolutional layers in identity_block, convolution kernel size is same in each convolutional layer
Respectively 1 × 1,3 × 3 and 1 × 1, after the output of the convolutional layer of the last layer is added with the input of identity_block
It is input to relu activation primitive layer.
Four branches connecting in Fig. 3 with Block5 are space characteristics extraction module, respectively to different spaces position
Feature extracts, and in more granularity neural networks, each space characteristics extraction module includes averagepooling average
The full articulamentum of pond layer, fc and softmax active coating, average pond layer, full articulamentum and softmax active coating are sequentially connected,
The feature being output in the branch of Block5 in foundation characteristic extraction module 12 is constantly learnt, four of Block5 connection
In branch, three characteristic patterns for branching into Block5 output above carry out the localized branches of cutting in the horizontal direction, and the 4th branches into
Global branch, by above-mentioned connection structure make three different localized branches by average pond layers, full articulamentum and
Softmax active coating can learn the notable feature to different zones, and global characteristics are arrived in study on global branch.It is understood that
It is that, according to the connection structure of Fig. 3, the size of the characteristic pattern exported in Block5 is 14 × 14, by this feature figure along level side
It is, for example, 14 × 5,14 × 4,14 × 5 to the characteristic pattern size of three branches for carrying out cutting acquisition, therefore at least needs pair
The characteristic pattern cutting of Block5 output is primary, obtains two localized branches and a global branch, obtains at least three spaces
Characteristic extracting module.
More granularity neural networks are different from existing common more granularity neural networks, multiple space characteristics extraction modules
Branch be completely independent, extract the feature of different parts respectively, and the stub network weight of more granularity neural networks is shared, drop
Low model complexity accelerates recognition speed, also, the overall situation for extracting in the form of global and more local mixings image is logical
With feature and local notable feature, global and local mutually supervision, promotion, so that the feature extracted is more steady.
Multiple outputs of more granularity neural networks are connect with the input of Fusion Features neural network, refering to Fig. 5, this feature
Fused neural network includes splicing layer 20, dimension transform layer 21, the second convolutional layer 22, third convolutional layer 23, Volume Four lamination 24
With the 5th convolutional layer 25, splice layer 20, dimension transform layer 21, the second convolutional layer 22, third convolutional layer 23,24 and of Volume Four lamination
5th convolutional layer 25 is sequentially connected, wherein splicing layer 20 is for splicing multiple outputs of more granularity neural networks, dimension
The output that transform layer 21 is used to splice layer 20 carries out dimension transformation.
Fig. 6 shows the connection knot of a kind of Fusion Features neural network and Classification Neural provided in this embodiment
Structure, four tunnel branches of more granularity neural networks carry out concatenate operation merging by splicing floor, and are transported by reshape
It calculates and dimension transformation is carried out to the output of splicing layer, according to connection structure shown in Fig. 3, dimension can be transformed to 4 by reshape ×
Vehicle money classification number × 1, wherein 4 rows respectively indicate 3 localized branches and 1 global branch, and every a line indicates different spaces region
On feature, each column indicate inhomogeneity another characteristic, therefore realize that each feature spatially decouples between classification.
It should be noted that as can be seen from Figure 6, this feature fused neural network, which is different from common symmetrical convolution kernel, to carry out
Fusion Features, the present embodiment using two groups of concatenated asymmetric convolution kernels convolutional network to each space branch and it is of all categories between
Feature merged, convolution kernel size is respectively 4 × 1 and 1 × 7, wherein it is different empty that 4 × 1 convolutional networks merge same category
Between feature between branch, 1 × 7 convolutional network merges the feature between the same space branch similar categorization, it is achieved that respectively
Feature Decoupled Fusion between space and classification, so that Fusion Features are highly efficient precisely, it is to be understood that Fusion Features nerve
The convolution kernel size of four convolutional layers can carry out adaptability according to the more granularity neural networks of upper level and actual conditions in network
Adjustment, the present embodiment are limited not to this, such as upper granularity neural network more than one layer uses a global branch and two offices
Part is propped up, then the convolution kernel size of 4 × 1 convolutional network can be changed to 3 × 1.
Feature of the features described above fused neural network by using the convolutional network blending image of asymmetric convolution kernel, parameter
Amount greatly reduces, and makes Fusion Features highly efficient, and finally obtained class categories feature is more significant, can effectively improve vehicle
The other accuracy of inscriptions.
The input of Classification Neural is connect with the output of Fusion Features neural network, according to the connection knot of Fig. 3 and Fig. 6
Structure, the output dimension of Fusion Features neural network are in 4 × vehicle money classification number × 8, Fig. 6, and Classification Neural passes through volume 4 × 1
Product network merges the feature for the space the category feature Zhong Ge branch that Fusion Features neural network exports, and obtains after fusion
Characteristic dimension is that 1 × vehicle money classification number × 1 obtains vehicle in the vehicle image of input and belong to each vehicle by softmax active coating
The probability of money classification, and it is exported.
It should be noted that similar to the same space branch due to passing through 1 × 7 convolutional network in Fusion Features neural network
Feature between classification is merged, therefore in Classification Neural output probability by the vehicle money classification with similar features
Together, such as a series of probability of similar vehicles in BMW car money can be arranged together and exported probabilistic.
It should also be noted that, more granularity neural networks, Fusion Features neural network, sorter network shown in attached drawing
Detailed construction in the size of each convolution kernel be only a kind of embodiment, be not construed as limitation of the invention.
In conclusion extracting each partial zones from the vehicle pictures of input by more granularity neural networks in above scheme
The notable feature in domain and global generic features, and the feature that more granularity neural networks are extracted is spatially divided into multiple parts point
The same row of branch and global branch, each branch indicates inhomogeneity another characteristic, therefore all features are between space and classification
Decoupling；Further, pass through the convolutional neural networks of the asymmetric convolution kernel of two-stage series connection in Fusion Features neural network point
Do not merge each space branching characteristic and it is of all categories between feature；Further, pass through an asymmetric convolution in Classification Neural
The convolutional neural networks of core merge each space branching characteristic and obtain all kinds of another characteristics, and are schemed by softmax active coating
Vehicle belongs to the probability of each vehicle money classification in piece, and the highest vehicle money of probability is determined as to the corresponding vehicle money of vehicle image of input,
Whole process recognition speed is fast, and is not influenced by vehicle color and shooting visual angle, for partial occlusion, motion blur and light
The factors such as dark influence, and can reach higher recognition correct rate.
It should be appreciated that above-mentioned to carry out vehicle inscriptions other service stage, input vehicle image then passes through more granularity nerve nets
Network, Fusion Features neural network, Classification Neural determine that the style of output vehicle need to pass through training before service stage
A large amount of training samples are concentrated to be trained each neural network, refering to Fig. 7, comprising:
Step 201: more granularity neural networks being trained using training sample on training set.
More granularity neural networks are trained on the training set containing several vehicle images, until training is stablized.
Step 202: the parameter in fixed more granularity neural networks to Fusion Features neural network and is divided using training sample
Neural network is trained.
After the completion of to more granularity neural metwork trainings, fixed parameter therein, since more granularity neural networks, feature are melted
Conjunction neural network, Classification Neural are sequentially connected, and the training sample in training set are inputted in more granularity neural networks, to spy
Sign fused neural network and Classification Neural are trained together, until Fusion Features neural network and Classification Neural are equal
Training is stablized, and then, can be finely adjusted by the learning rate of very little to whole neural networks, to reach more preferably Classification and Identification effect
Fruit.
Second embodiment
The present embodiment provides a kind of vehicle money identifying systems, refering to Fig. 8, comprising:
Characteristic extracting module 30, for by more granularity neural networks, extracting the global characteristics of vehicle image and multiple
Local feature；
Fusion Features module 31, for by Fusion Features neural network to the global characteristics of the vehicle image with
And multiple local features are merged, and the corresponding category feature of the vehicle image is obtained；
Categorization module 32, for obtaining the vehicle image by Classification Neural and belonging to according to the category feature
The probability of each vehicle money classification；
The highest vehicle money of probability is determined as the corresponding vehicle money of the vehicle image by vehicle money determining module 33.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
Specific work process, no longer can excessively be repeated herein with reference to the corresponding process in preceding method.
The present embodiment also provides a kind of electronic equipment, comprising: processor and memory, which is stored with processor can
The machine readable instructions of execution execute method described in first embodiment when machine readable instructions are executed by processor.
The present embodiment also provides a kind of storage medium, and computer program, computer program quilt are stored on the storage medium
Method described in first embodiment is executed when processor is run.
In embodiment provided herein, it should be understood that disclosed system and method, it can also be by other
Mode realize.System embodiment described above is only schematical, for example, the flow chart and block diagram in attached drawing are shown
The system of multiple embodiments according to the present invention, the architectural framework in the cards of method and computer program product, function
And operation.In this regard, each box in flowchart or block diagram can represent one of a module, section or code
Point, a part of the module, section or code includes one or more for implementing the specified logical function executable
Instruction.It should also be noted that function marked in the box can also be attached to be different from some implementations as replacement
The sequence marked in figure occurs.For example, two continuous boxes can actually be basically executed in parallel, they sometimes may be used
To execute in the opposite order, this depends on the function involved.It is also noted that each of block diagram and or flow chart
The combination of box in box and block diagram and or flow chart can be based on the defined function of execution or the dedicated of movement
The system of hardware is realized, or can be realized using a combination of dedicated hardware and computer instructions.
In addition, each functional module in each embodiment of the present invention can integrate one independent portion of formation together
Point, it is also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It, can be with if the function is realized and when sold or used as an independent product in the form of software function module
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, laptop, server or network equipment etc.) execute the whole of each embodiment the method for the present invention
Or part steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.It should be noted that, in this document, relational terms such as first and second and the like are used merely to one
A entity or operation with another entity or operate distinguish, without necessarily requiring or implying these entities or operation it
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant are intended to
Cover non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or setting
Standby intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in the process, method, article or apparatus that includes the element.
The foregoing is merely alternative embodiments of the invention, are not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of vehicle money recognition methods characterized by comprising
Using more granularity neural networks, the global characteristics and multiple local features of vehicle image are extracted；
The global characteristics of the vehicle image and multiple local features are merged using Fusion Features neural network,
Obtain the corresponding category feature of the vehicle image；
According to the category feature, the probability that the vehicle image belongs to each vehicle money classification is obtained using Classification Neural；
The highest vehicle money of probability is determined as the corresponding vehicle money of the vehicle image.
2. the method according to claim 1, wherein more granularity neural networks include the first convolutional layer, most
Great Chiization layer, at least one foundation characteristic extraction module and at least three space characteristics extraction modules, first convolutional layer,
Maximum pond layer, at least one described foundation characteristic extraction module are sequentially connected, at least one described foundation characteristic extracts
The output of module is connect at least three space characteristics extraction module respectively；
The foundation characteristic extraction module is used to extract the foundation characteristic in vehicle image, each space characteristics extraction module
For learning corresponding global characteristics or local feature, the foundation characteristic extraction module includes conv_block and at least one
Identity_block, the conv_block and at least one described identity_block are sequentially connected.
3. according to the method described in claim 2, it is characterized in that, each space characteristics extraction module includes average pond
Layer, full articulamentum and softmax active coating, the average pond layer, the full articulamentum and the softmax active coating are successively
Connection.
4. according to the method described in claim 3, it is characterized in that, the Fusion Features neural network includes splicing layer, dimension
Transform layer, the second convolutional layer, third convolutional layer, Volume Four lamination and the 5th convolutional layer, the splicing layer, dimension transformation
Layer, second convolutional layer, the third convolutional layer, the Volume Four lamination and the 5th convolutional layer are sequentially connected, described
For splicing multiple outputs of more granularity neural networks, the dimension transform layer is used for the splicing splicing layer
The output of layer carries out dimension transformation.
5. according to the method described in claim 4, it is characterized in that, the convolution kernel of second convolutional layer and Volume Four product
The convolution kernel of layer is identical, and the convolution kernel of the third convolutional layer is identical as the convolution kernel of the 5th convolutional layer, the volume Two
The convolution kernel of lamination, the convolution kernel of the third convolutional layer, the convolution kernel of the Volume Four lamination and the 5th convolutional layer
Convolution kernel is asymmetric convolution kernel.
6. according to the method described in claim 5, utilizing classification nerve net it is characterized in that, described according to the category feature
Network obtains the probability that the vehicle image belongs to each vehicle money classification, specifically:
The Classification Neural merges the feature of the space category feature Zhong Ge branch by the 6th convolutional layer, passes through
Softmax active coating exports the probability that the vehicle image belongs to each vehicle money classification.
7. method according to claim 1 to 6, which is characterized in that utilizing more granularity neural networks, extracting
Before the global characteristics of vehicle image and multiple local features, the method also includes:
More granularity neural networks are trained using the training sample in training set until meeting training termination condition.
8. the method according to the description of claim 7 is characterized in that in the training sample using in training set to described more
After granularity neural network is trained until meeting training termination condition, the method also includes:
Parameter in fixed more granularity neural networks, using the training sample to the Fusion Features neural network and institute
Classification Neural is stated to be trained.
9. a kind of vehicle money identifying system characterized by comprising
Characteristic extracting module, for by more granularity neural networks, global characteristics and the multiple parts for extracting vehicle image to be special
Sign；
Fusion Features module, for by Fusion Features neural network to the global characteristics of the vehicle image and multiple
Local feature is merged, and the corresponding category feature of the vehicle image is obtained；
Categorization module, for obtaining the vehicle image by Classification Neural and belonging to each vehicle money according to the category feature
The probability of classification；
The highest vehicle money of probability is determined as the corresponding vehicle money of the vehicle image by vehicle money determining module.
10. a kind of storage medium, which is characterized in that be stored with computer program, the computer program on the storage medium
The method according to claim 1 is executed when being run by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811268068.2A CN109359696B (en) | 2018-10-29 | 2018-10-29 | Vehicle money identification method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811268068.2A CN109359696B (en) | 2018-10-29 | 2018-10-29 | Vehicle money identification method, system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109359696A true CN109359696A (en) | 2019-02-19 |
| CN109359696B CN109359696B (en) | 2021-04-02 |
Family
ID=65347074
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811268068.2A Active CN109359696B (en) | 2018-10-29 | 2018-10-29 | Vehicle money identification method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359696B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210750A (en) * | 2019-05-29 | 2019-09-06 | 北京天正聚合科技有限公司 | A kind of method, apparatus, electronic equipment and storage medium identifying Shopping Guide's business |
| CN110347164A (en) * | 2019-08-08 | 2019-10-18 | 北京云迹科技有限公司 | A kind of speed adjusting method, device and storage medium |
| CN110516583A (en) * | 2019-08-21 | 2019-11-29 | 中科视语(北京)科技有限公司 | A kind of vehicle recognition methods, system, equipment and medium again |
| CN110795580A (en) * | 2019-10-23 | 2020-02-14 | 武汉理工大学 | Vehicle weight recognition method based on space-time constraint model optimization |
| CN110969610A (en) * | 2019-12-03 | 2020-04-07 | 杭州天铂云科光电科技有限公司 | Power equipment infrared chart identification method and system based on deep learning |
| CN111461181A (en) * | 2020-03-16 | 2020-07-28 | 北京邮电大学 | Vehicle fine-grained classification method and device |
| CN111598131A (en) * | 2020-04-17 | 2020-08-28 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN111611414A (en) * | 2019-02-22 | 2020-09-01 | 杭州海康威视数字技术股份有限公司 | Vehicle retrieval method, device and storage medium |
| CN111832369A (en) * | 2019-04-23 | 2020-10-27 | ***通信有限公司研究院 | Image identification method and device and electronic equipment |
| CN112101246A (en) * | 2020-09-18 | 2020-12-18 | 济南博观智能科技有限公司 | Vehicle identification method, device, equipment and medium |
| CN112132031A (en) * | 2020-09-23 | 2020-12-25 | 平安国际智慧城市科技股份有限公司 | Vehicle money identification method and device, electronic equipment and storage medium |
| CN112446402A (en) * | 2019-09-03 | 2021-03-05 | 顺丰科技有限公司 | Loading rate identification method and device, computer equipment and storage medium |
| CN112766407A (en) * | 2021-01-29 | 2021-05-07 | 北京达佳互联信息技术有限公司 | Image recognition method, device and storage medium |
| CN112990048A (en) * | 2021-03-26 | 2021-06-18 | 中科视语(北京)科技有限公司 | Vehicle pattern recognition method and device |
| CN113177502A (en) * | 2021-05-13 | 2021-07-27 | 北京罗克维尔斯科技有限公司 | Looking-around obstacle detection method, device, medium, vehicle-mounted system and vehicle |
| CN113920313A (en) * | 2021-09-29 | 2022-01-11 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN117173100A (en) * | 2023-07-31 | 2023-12-05 | 东莞市言科新能源有限公司 | Polymer lithium ion battery production control system and method thereof |
| WO2024001653A1 (en) * | 2022-06-30 | 2024-01-04 | 京东方科技集团股份有限公司 | Feature extraction method and apparatus, storage medium, and electronic device |
| WO2024066697A1 (en) * | 2022-09-26 | 2024-04-04 | 腾讯科技（深圳）有限公司 | Image processing method and related apparatus |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7529391B2 (en) * | 2005-12-29 | 2009-05-05 | Microsoft Corporation | Signature verification |
| CN105488517A (en) * | 2015-11-30 | 2016-04-13 | 杭州全实鹰科技有限公司 | Vehicle brand model identification method based on deep learning |
| CN106548145A (en) * | 2016-10-31 | 2017-03-29 | 北京小米移动软件有限公司 | Image-recognizing method and device |
| CN107330463A (en) * | 2017-06-29 | 2017-11-07 | 南京信息工程大学 | Model recognizing method based on CNN multiple features combinings and many nuclear sparse expressions |
| US20170351941A1 (en) * | 2016-06-03 | 2017-12-07 | Miovision Technologies Incorporated | System and Method for Performing Saliency Detection Using Deep Active Contours |
| CN108021933A (en) * | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| 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 |
-
2018
- 2018-10-29 CN CN201811268068.2A patent/CN109359696B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7529391B2 (en) * | 2005-12-29 | 2009-05-05 | Microsoft Corporation | Signature verification |
| CN105488517A (en) * | 2015-11-30 | 2016-04-13 | 杭州全实鹰科技有限公司 | Vehicle brand model identification method based on deep learning |
| US20170351941A1 (en) * | 2016-06-03 | 2017-12-07 | Miovision Technologies Incorporated | System and Method for Performing Saliency Detection Using Deep Active Contours |
| CN106548145A (en) * | 2016-10-31 | 2017-03-29 | 北京小米移动软件有限公司 | Image-recognizing method and device |
| CN107330463A (en) * | 2017-06-29 | 2017-11-07 | 南京信息工程大学 | Model recognizing method based on CNN multiple features combinings and many nuclear sparse expressions |
| CN108021933A (en) * | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| 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 |
Non-Patent Citations (3)
| Title |
|---|
| SHUAI ZHAO, ET AL.: "Research on the Recognition of Car Models Based on Deep-Learning Networks", 《PROCEEDINGS OF THE 2017 ASIA-PACIFIC COMPUTER SCIENCE AND APPLICATION CONFERENCE》 * |
| 孙晔: "基于卷积神经网络的车标识别", 《现代计算机》 * |
| 甘凯令，等: "融合整体与局部特征的车辆型号识别方法", 《现代电子技术》 * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111611414A (en) * | 2019-02-22 | 2020-09-01 | 杭州海康威视数字技术股份有限公司 | Vehicle retrieval method, device and storage medium |
| CN111611414B (en) * | 2019-02-22 | 2023-10-24 | 杭州海康威视数字技术股份有限公司 | Vehicle searching method, device and storage medium |
| CN111832369A (en) * | 2019-04-23 | 2020-10-27 | ***通信有限公司研究院 | Image identification method and device and electronic equipment |
| CN110210750A (en) * | 2019-05-29 | 2019-09-06 | 北京天正聚合科技有限公司 | A kind of method, apparatus, electronic equipment and storage medium identifying Shopping Guide's business |
| CN110347164A (en) * | 2019-08-08 | 2019-10-18 | 北京云迹科技有限公司 | A kind of speed adjusting method, device and storage medium |
| CN110516583A (en) * | 2019-08-21 | 2019-11-29 | 中科视语(北京)科技有限公司 | A kind of vehicle recognition methods, system, equipment and medium again |
| CN112446402B (en) * | 2019-09-03 | 2024-03-26 | 顺丰科技有限公司 | Loading rate identification method, loading rate identification device, computer equipment and storage medium |
| CN112446402A (en) * | 2019-09-03 | 2021-03-05 | 顺丰科技有限公司 | Loading rate identification method and device, computer equipment and storage medium |
| CN110795580B (en) * | 2019-10-23 | 2023-12-08 | 武汉理工大学 | Vehicle weight identification method based on space-time constraint model optimization |
| CN110795580A (en) * | 2019-10-23 | 2020-02-14 | 武汉理工大学 | Vehicle weight recognition method based on space-time constraint model optimization |
| CN110969610A (en) * | 2019-12-03 | 2020-04-07 | 杭州天铂云科光电科技有限公司 | Power equipment infrared chart identification method and system based on deep learning |
| CN111461181A (en) * | 2020-03-16 | 2020-07-28 | 北京邮电大学 | Vehicle fine-grained classification method and device |
| CN111598131A (en) * | 2020-04-17 | 2020-08-28 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| US11727676B2 (en) | 2020-04-17 | 2023-08-15 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Feature fusion method and apparatus for image processing, electronic device and storage medium |
| CN111598131B (en) * | 2020-04-17 | 2023-08-25 | 北京百度网讯科技有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN112101246A (en) * | 2020-09-18 | 2020-12-18 | 济南博观智能科技有限公司 | Vehicle identification method, device, equipment and medium |
| CN112132031A (en) * | 2020-09-23 | 2020-12-25 | 平安国际智慧城市科技股份有限公司 | Vehicle money identification method and device, electronic equipment and storage medium |
| CN112132031B (en) * | 2020-09-23 | 2024-04-16 | 平安国际智慧城市科技股份有限公司 | Vehicle style identification method and device, electronic equipment and storage medium |
| CN112766407B (en) * | 2021-01-29 | 2023-12-05 | 北京达佳互联信息技术有限公司 | Image recognition method, device and storage medium |
| CN112766407A (en) * | 2021-01-29 | 2021-05-07 | 北京达佳互联信息技术有限公司 | Image recognition method, device and storage medium |
| CN112990048B (en) * | 2021-03-26 | 2021-11-23 | 中科视语(北京)科技有限公司 | Vehicle pattern recognition method and device |
| CN112990048A (en) * | 2021-03-26 | 2021-06-18 | 中科视语(北京)科技有限公司 | Vehicle pattern recognition method and device |
| CN113177502A (en) * | 2021-05-13 | 2021-07-27 | 北京罗克维尔斯科技有限公司 | Looking-around obstacle detection method, device, medium, vehicle-mounted system and vehicle |
| CN113920313A (en) * | 2021-09-29 | 2022-01-11 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| WO2024001653A1 (en) * | 2022-06-30 | 2024-01-04 | 京东方科技集团股份有限公司 | Feature extraction method and apparatus, storage medium, and electronic device |
| WO2024066697A1 (en) * | 2022-09-26 | 2024-04-04 | 腾讯科技（深圳）有限公司 | Image processing method and related apparatus |
| CN117173100A (en) * | 2023-07-31 | 2023-12-05 | 东莞市言科新能源有限公司 | Polymer lithium ion battery production control system and method thereof |
| CN117173100B (en) * | 2023-07-31 | 2024-04-05 | 东莞市言科新能源有限公司 | Polymer lithium ion battery production control system and method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109359696B (en) | 2021-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109359696A (en) | A kind of vehicle money recognition methods, system and storage medium | |
| CN105574550B (en) | A kind of vehicle identification method and device | |
| CN109740413A (en) | Pedestrian recognition methods, device, computer equipment and computer storage medium again | |
| CN109145766A (en) | Model training method, device, recognition methods, electronic equipment and storage medium | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN105069448A (en) | True and false face identification method and device | |
| US11640714B2 (en) | Video panoptic segmentation | |
| CN105654066A (en) | Vehicle identification method and device | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN108197326A (en) | A kind of vehicle retrieval method and device, electronic equipment, storage medium | |
| CN108564673A (en) | A kind of check class attendance method and system based on Global Face identification | |
| CN101142584A (en) | Method for facial features detection | |
| CN109376631A (en) | A kind of winding detection method and device neural network based | |
| Funk et al. | Beyond planar symmetry: Modeling human perception of reflection and rotation symmetries in the wild | |
| CN106408037A (en) | Image recognition method and apparatus | |
| CN111274886B (en) | Deep learning-based pedestrian red light running illegal behavior analysis method and system | |
| Jain et al. | Performance analysis of object detection and tracking algorithms for traffic surveillance applications using neural networks | |
| CN109902202A (en) | A kind of video classification methods and device | |
| CN112529005B (en) | Target detection method based on semantic feature consistency supervision pyramid network | |
| CN110532855A (en) | Natural scene certificate image character recognition method based on deep learning | |
| CN112257665A (en) | Image content recognition method, image recognition model training method, and medium | |
| Xing et al. | Traffic sign recognition using guided image filtering | |
| CN112488229A (en) | Domain self-adaptive unsupervised target detection method based on feature separation and alignment | |
| Wechsler et al. | Automatic video-based person authentication using the RBF network | |
| Li et al. | Image manipulation localization using attentional cross-domain CNN features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 401122 5 stories, Block 106, West Jinkai Avenue, Yubei District, Chongqing Applicant after: Chongqing Zhongke Yuncong Technology Co., Ltd. Address before: 400 000 5 stories of Block 106, West Jinkai Avenue, Yubei District, Chongqing Applicant before: CHONGQING ZHONGKE YUNCONG TECHNOLOGY CO., LTD. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |