CN110175622A - The vehicle part recognition methods of convolutional neural networks based on symbiosis and system - Google Patents
The vehicle part recognition methods of convolutional neural networks based on symbiosis and system Download PDFInfo
- Publication number
- CN110175622A CN110175622A CN201910269801.0A CN201910269801A CN110175622A CN 110175622 A CN110175622 A CN 110175622A CN 201910269801 A CN201910269801 A CN 201910269801A CN 110175622 A CN110175622 A CN 110175622A
- Authority
- CN
- China
- Prior art keywords
- symbiosis
- vehicle part
- convolutional neural
- neural networks
- recognition methods
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
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)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle part recognition methods of convolutional neural networks based on symbiosis and systems, belong to mode identification technology.It include: using vehicle part to convolutional neural networks of the training dataset training based on symbiosis, it is composed in series by Concat layers with conventional convolution neural network based on the convolutional neural networks of symbiosis, Concat layers for the picture to input to progress channel merging;Using the trained convolutional neural networks based on symbiosis, treats and survey vehicle part to identifying, obtain the recognition result of various vehicle components to be measured.The present invention merges picture by channel by Concat layers, when subsequent network carries out convolution operation, can merge the information in the corresponding different channels of two components, while extracting the feature of two pictures, symbiosis is arrived by preferably study.The introducing of symbiosis is added to additional information, in a disguised form increases trained data volume, improves the recognition accuracy of two base parts.
Description
Technical field
The invention belongs to mode identification technologies, more particularly, to a kind of convolutional Neural net based on symbiosis
The vehicle part recognition methods of network and system.
Background technique
Vehicle part identification be intelligent transportation system important component, need to identify car light, rearview mirror, logo,
The subclass of air inlet grill and windshield identifies that is, by identifying the concrete shape of car light, rearview mirror, logo, windshield
The title of logo.
Vehicle part recognizer substantially belongs to sorting algorithm, is divided into using the algorithm of conventional machines study and based on volume
The algorithm of product neural network.Method based on machine learning is divided into two steps, and the first step is to extract the characteristics of image of component to be identified,
Second step is to go to classify to the feature of extraction using classifier, finally obtains the subclass of component to be identified, common methods
Having morphological feature to add decision tree, Chebyshev, bending moment does not add SVM etc..Algorithm based on convolutional neural networks is usual
It is directly to treat identification component using networks such as VGG, ResNet to classify.
However above-mentioned algorithm all only focuses on the identification in single component, not in view of the connection between different components, leads
Cause the recognition accuracy of each base part not high.
Summary of the invention
In view of the drawbacks of the prior art, it is an object of the invention to solve the recognition methods of prior art vehicle part not examine
Consider the technical problem that symbiosis causes each base part recognition accuracy low.
To achieve the above object, in a first aspect, the embodiment of the invention provides a kind of convolutional Neurals based on symbiosis
The vehicle part recognition methods of network, method includes the following steps:
S1. described based on altogether using vehicle part to convolutional neural networks of the training dataset training based on symbiosis
The convolutional neural networks of raw relationship are composed in series by Concat layers with conventional convolution neural network, and described Concat layers for defeated
The picture entered is to progress channel merging;
S2. the trained convolutional neural networks based on symbiosis are used, treats and surveys vehicle part to identifying, obtain
To the recognition result of various vehicle components to be measured.
Specifically, rearview mirror, car light, logo, four base part of windshield are chosen as vehicle part and identifies object, by institute
State four base parts and carry out combination of two forming member pair, obtain 6 vehicle parts pair:<rearview mirror, car light>,<rearview mirror, logo
>,<rearview mirror, windshield>,<car light, logo>,<car light, windshield>,<logo, windshield>.
Specifically, the conventional convolution neural network is ResNet VGG network.
Specifically, two n-channel pictures after Concat layers, are becoming the tensor that a port number is 2n.
Specifically, mutual information is used to measure the power of symbiosis between two base part of component centering, the bigger table of association relationship
Bright symbiosis is stronger, and symbiosis is stronger, and vehicle part recognition accuracy is higher.
Second aspect, the embodiment of the invention provides a kind of knowledges of the vehicle part of convolutional neural networks based on symbiosis
Other system, the vehicle part identifying system is using the convolutional neural networks based on symbiosis described in above-mentioned first aspect
Vehicle part recognition methods.
The third aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage mediums
Computer program is stored in matter, which realizes described in above-mentioned first aspect when being executed by processor based on symbiosis
The vehicle part recognition methods of the convolutional neural networks of relationship.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, have below beneficial to effect
Fruit:
The present invention proposes a kind of convolutional neural networks based on symbiosis, after Concat layers merge picture by channel, after
When continuous network carries out convolution operation, the information in the corresponding different channels of two components can be merged, i.e., network can mention simultaneously
The feature for taking two pictures arrives symbiosis by preferably study.The introducing of symbiosis is added to additional letter
Breath, in a disguised form increases trained data volume.Data due to being input to the convolutional neural networks based on symbiosis contain altogether
Raw relationship, during subsequent training study, which can learn the symbiosis to two base parts automatically, and in reality
In use, improving the recognition accuracy of these two types of components using the symbiosis learnt.
Detailed description of the invention
Fig. 1 is a kind of vehicle part identification side of the convolutional neural networks based on symbiosis provided in an embodiment of the present invention
Method flow chart;
Fig. 2 is the convolutional neural networks structural schematic diagram provided in an embodiment of the present invention based on symbiosis.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
As shown in Figure 1, the present invention proposes a kind of vehicle part recognition methods of convolutional neural networks based on symbiosis,
Method includes the following steps:
S1. described based on altogether using vehicle part to convolutional neural networks of the training dataset training based on symbiosis
The convolutional neural networks of raw relationship are composed in series by Concat layers with conventional convolution neural network, and described Concat layers for defeated
The picture entered is to progress channel merging;
S2. the trained convolutional neural networks based on symbiosis are used, treats and surveys vehicle part to identifying, obtain
To the recognition result of various vehicle components to be measured.
Step S1. is using vehicle part to convolutional neural networks of the training dataset training based on symbiosis, the base
It is composed in series by Concat layers with conventional convolution neural network in the convolutional neural networks of symbiosis, described Concat layers is used for
To the picture of input to carrying out channel merging.
The present invention chooses rearview mirror, car light, logo, four base part of windshield as identification object, by rearview mirror, vehicle
Lamp, logo, four base part of windshield carry out combination of two forming member pair, are obtained 6 components pair:<rearview mirror, and car light>,
<rearview mirror, logo>,<rearview mirror, windshield>,<car light, logo>,<car light, windshield>,<logo, windshield>.
In the present embodiment, vehicle part includes the vehicle data set of 90224 positive faces of vehicle from one to data set.
Vehicle data source vehicle captured by the camera of crossroad monitoring road.Most of shooting time is daytime, is clapped
The positive face that angle is vehicle is taken the photograph, the vehicle of all common types is covered.Shooting image is colour, and resolution ratio is about 800 ×
700.By the cutting to vehicle data set, vehicle part is obtained to data set.
In a practical situation, the classification of each component of a vehicle is not arbitrarily to arrange in pairs or groups.For example, specific shape
Car light can only appear on the vehicle comprising specific shape rearview mirror.More specifically, logo is the corresponding air inlet grill of automobile of BMW
Shape is essentially all vertical grid.Make the classification of another base part also general since the classification of a base part is determining
The constraint relationship that rate is determined, existing the constraint relationship is referred to as " symbiosis " between this component.
It can use symbiosis the invention proposes one to improve the network of each component subclass recognition accuracy.Institute
Network is stated before the convolutional layer of conventional convolution neural network (ResNet or VGG), increases by one Concat layers, for defeated
The picture entered is to progress channel merging.It, can be by different channels when subsequent network carries out convolution operation after merging picture by channel
Information is merged, i.e., network can extract the feature of two pictures simultaneously, and symbiosis is arrived by preferably study.By
Symbiosis is contained in the data for being input to the convolutional neural networks based on symbiosis, in the process of subsequent training study
In, which can learn the symbiosis to two base parts automatically, and in actual use, be closed using the symbiosis learnt
The recognition accuracy of these two types of components improves in system.
As shown in Fig. 2, the input of the convolutional neural networks based on symbiosis is vehicle part pair, output is each part of vehicle
Part generic, the network can identify two base parts of vehicle simultaneously.Concat layer is used to merge input by channel
Picture pair, so that two n-channel pictures are becoming the tensor that a port number is 2n after Concat layer.
Training sample is sent into network, corresponding hyper parameter (learning rate, the number of iterations, the size of batch) is set, to net
Network is trained.The loss function for identifying two base parts is respectively l1And l2, total loss function is L, utilizes parameter lambda1And λ2Come
Adjust the weight of the two, i.e. L=λ1l1+λ2l2。
Step S2. uses the trained convolutional neural networks based on symbiosis, treats and surveys vehicle part to knowing
Not, the recognition result of various vehicle components to be measured is obtained.
For example, vehicle part to be measured is to for<rearview mirror, car light>, the recognition result of obtained various vehicle components to be measured
Subcategory1 are as follows: rearview mirror is one in { lateral rectangle, triangle is round, and square, longitudinal rectangle can not identify }
Kind, recognition result subcategory2 are as follows: car light is { rectangle, irregular figure, triangle, quadrangle is round, can not identify }
One of.
The vehicle part recognition methods and existing skill of convolutional neural networks more proposed by the present invention based on symbiosis
Vehicle part recognition methods in art based on ResNet, comparison result are as shown in table 1.
The calculation formula of accuracy rate Acc is as follows:
Wherein, tp is the sample number that positive sample is predicted as to positive example, and fp is the sample number that negative sample is predicted as to positive example.
Table 1
It can be seen that the recognition accuracy of other component has promotion in addition to car light.This also experimentally demonstrates different portions
Symbiosis between part can promote recognition accuracy, i.e., algorithm proposed by the present invention has significantly in terms of vehicle part identification
Promote effect.
But the recognition accuracy of car light is declined, and the mutual information for observing car light and other component can be found that (table 2), with
The related association relationship of car light is smaller.Symbiosis i.e. between car light and other component is weaker.E-learning is less than enough
Information related with symbiosis, introduces some irrelevant informations instead, so that recognition accuracy declines.
Mutual information (Mutual Information) is a kind of useful measure information in information theory, it can regard one as
The information content about another stochastic variable for including in a stochastic variable, or perhaps a stochastic variable is due to known another
A stochastic variable and the uncertainty of reduction.For two stochastic variables, the calculation formula of mutual information is as follows:
With component to for<rearview mirror, car light>for, X indicates type set { lateral rectangle, triangle, the circle of rearview mirror
Shape, square, longitudinal rectangle can not identify }, Y indicates type set { rectangle, irregular figure, triangle, four sides of car light
Shape, it is round, can not identify, p (x) indicates that rearview mirror is certain types of ratio, and p (y) indicates that car light is certain types of ratio
Example.
Table 2 gives the mutual information of each component pair, and mutual information is used to measure the strong of symbiosis between two component of component centering
Weak, association relationship shows that more greatly symbiosis is stronger.
Table 2
More than, the only preferable specific embodiment of the application, but the protection scope of the application is not limited thereto, and it is any
Within the technical scope of the present application, any changes or substitutions that can be easily thought of by those familiar with the art, all answers
Cover within the scope of protection of this application.Therefore, the protection scope of the application should be subject to the protection scope in claims.
Claims (7)
1. a kind of vehicle part recognition methods of the convolutional neural networks based on symbiosis, which is characterized in that this method includes
Following steps:
S1. described to be closed based on symbiosis using vehicle part to convolutional neural networks of the training dataset training based on symbiosis
The convolutional neural networks of system are composed in series by Concat layers with conventional convolution neural network, and described Concat layers for input
Picture is to progress channel merging;
S2. use the trained convolutional neural networks based on symbiosis, treat survey vehicle part to identifying, obtain to
Survey the recognition result of various vehicle components.
2. vehicle part recognition methods as described in claim 1, which is characterized in that choose rearview mirror, car light, logo, keep out the wind
Four base part of glass identifies object as vehicle part, and four base part is carried out combination of two forming member pair, obtains 6
Vehicle part pair:<rearview mirror, car light>,<rearview mirror, logo>,<rearview mirror, windshield>,<car light, logo>,<car light, gear
Wind glass>,<logo, windshield>.
3. vehicle part recognition methods as described in claim 1, which is characterized in that the conventional convolution neural network is
ResNet VGG network.
4. vehicle part recognition methods as described in claim 1, which is characterized in that two n-channel pictures are passing through Concat
After layer, become the tensor that a port number is 2n.
5. vehicle part recognition methods as described in claim 1, which is characterized in that mutual information is for measuring two class of component centering
The power of symbiosis between component, association relationship show that more greatly symbiosis is stronger, and symbiosis is stronger, and vehicle part identification is quasi-
True rate is higher.
6. a kind of vehicle part identifying system of the convolutional neural networks based on symbiosis, which is characterized in that the vehicle portion
Part identifying system uses the vehicle part such as the convolutional neural networks described in any one of claim 1 to 5 based on symbiosis
Recognition methods.
7. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program, when the computer program is executed by processor realize as it is described in any one of claim 1 to 5 based on symbiosis
The vehicle part recognition methods of convolutional neural networks.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910269801.0A CN110175622B (en) | 2019-04-04 | 2019-04-04 | Vehicle component identification method and system based on convolutional neural network of symbiotic relationship |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910269801.0A CN110175622B (en) | 2019-04-04 | 2019-04-04 | Vehicle component identification method and system based on convolutional neural network of symbiotic relationship |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110175622A true CN110175622A (en) | 2019-08-27 |
CN110175622B CN110175622B (en) | 2021-10-08 |
Family
ID=67689412
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910269801.0A Expired - Fee Related CN110175622B (en) | 2019-04-04 | 2019-04-04 | Vehicle component identification method and system based on convolutional neural network of symbiotic relationship |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110175622B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111402200A (en) * | 2020-02-18 | 2020-07-10 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
CN115808944A (en) * | 2023-02-09 | 2023-03-17 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106384100A (en) * | 2016-09-28 | 2017-02-08 | 武汉大学 | Component-based fine vehicle model recognition method |
CN106529446A (en) * | 2016-10-27 | 2017-03-22 | 桂林电子科技大学 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
CN106971187A (en) * | 2017-04-12 | 2017-07-21 | 华中科技大学 | A kind of vehicle part detection method and system based on vehicle characteristics point |
US20180046896A1 (en) * | 2016-08-12 | 2018-02-15 | DeePhi Technology Co., Ltd. | Method and device for quantizing complex artificial neural network |
-
2019
- 2019-04-04 CN CN201910269801.0A patent/CN110175622B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180046896A1 (en) * | 2016-08-12 | 2018-02-15 | DeePhi Technology Co., Ltd. | Method and device for quantizing complex artificial neural network |
CN106384100A (en) * | 2016-09-28 | 2017-02-08 | 武汉大学 | Component-based fine vehicle model recognition method |
CN106529446A (en) * | 2016-10-27 | 2017-03-22 | 桂林电子科技大学 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
CN106971187A (en) * | 2017-04-12 | 2017-07-21 | 华中科技大学 | A kind of vehicle part detection method and system based on vehicle characteristics point |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111402200A (en) * | 2020-02-18 | 2020-07-10 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
CN111402200B (en) * | 2020-02-18 | 2021-12-21 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
CN115808944A (en) * | 2023-02-09 | 2023-03-17 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
CN115808944B (en) * | 2023-02-09 | 2023-06-02 | 国能大渡河枕头坝发电有限公司 | Stator temperature rise test control method |
Also Published As
Publication number | Publication date |
---|---|
CN110175622B (en) | 2021-10-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109816024B (en) | Real-time vehicle logo detection method based on multi-scale feature fusion and DCNN | |
CN112818862B (en) | Face tampering detection method and system based on multi-source clues and mixed attention | |
CN108549926A (en) | A kind of deep neural network and training method for refining identification vehicle attribute | |
CN111079674B (en) | Target detection method based on global and local information fusion | |
CN110222604B (en) | Target identification method and device based on shared convolutional neural network | |
CN107992819B (en) | Method and device for determining vehicle attribute structural features | |
CN108090411A (en) | Traffic lights detection and classification are carried out using computer vision and deep learning | |
CN112163511B (en) | Method for identifying authenticity of image | |
CN110852316A (en) | Image tampering detection and positioning method adopting convolution network with dense structure | |
CN109948616A (en) | Image detecting method, device, electronic equipment and computer readable storage medium | |
CN110929635B (en) | False face video detection method and system based on face cross-over ratio under trust mechanism | |
CN111242955B (en) | Road surface crack image segmentation method based on full convolution neural network | |
CN110826411B (en) | Vehicle target rapid identification method based on unmanned aerial vehicle image | |
CN112949578B (en) | Vehicle lamp state identification method, device, equipment and storage medium | |
CN111832461A (en) | Non-motor vehicle riding personnel helmet wearing detection method based on video stream | |
CN110599453A (en) | Panel defect detection method and device based on image fusion and equipment terminal | |
CN106407951A (en) | Monocular vision-based nighttime front vehicle detection method | |
CN110175622A (en) | The vehicle part recognition methods of convolutional neural networks based on symbiosis and system | |
CN109284759A (en) | One kind being based on the magic square color identification method of support vector machines (svm) | |
CN117765373A (en) | Lightweight road crack detection method and system with self-adaptive crack size | |
CN117011563A (en) | Road damage inspection cross-domain detection method and system based on semi-supervised federal learning | |
CN112926595B (en) | Training device of deep learning neural network model, target detection system and method | |
Karthigadevi et al. | Speed limit signboard recognition and audio alert for driver using CNN | |
CN109816022A (en) | A kind of image-recognizing method based on three decisions and CNN | |
CN110490053B (en) | Human face attribute identification method based on trinocular camera depth estimation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211008 |
|
CF01 | Termination of patent right due to non-payment of annual fee |