CN110991349A - Lightweight vehicle attribute identification method based on metric learning - Google Patents
Lightweight vehicle attribute identification method based on metric learning Download PDFInfo
- Publication number
- CN110991349A CN110991349A CN201911234612.6A CN201911234612A CN110991349A CN 110991349 A CN110991349 A CN 110991349A CN 201911234612 A CN201911234612 A CN 201911234612A CN 110991349 A CN110991349 A CN 110991349A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- loss function
- vehicle attribute
- sample
- lightweight
- 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
Images
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
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
- G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a lightweight vehicle attribute identification method based on metric learning, which comprises the following steps: training a lightweight convolutional neural network according to the multi-angle vehicle sample image to obtain a vehicle attribute identification model; acquiring a vehicle image, and inputting the vehicle attribute identification model to acquire a vehicle attribute; wherein the vehicle attributes comprise vehicle type, color, brand information of the vehicle; the vehicle attribute identification method provided by the invention can directly extract the depth characteristics by using the lightweight convolution neural network after obtaining the vehicle image to be identified, and can realize the rapid and accurate identification of vehicle types, colors, brands and other vehicle attributes by using only one neural network.
Description
Technical Field
The invention relates to the field of vehicle identification, in particular to a lightweight vehicle attribute identification method based on metric learning.
Background
With the rapid development of national economy, the increase of the number of vehicles brings great challenges to traffic supervision, and an intelligent traffic security system is considered to be a most effective solution for relieving traffic pressure. The vehicle attribute recognition is an important technology in the field of intelligent traffic security, realizes attribute recognition for vehicles in mass monitoring data, is beneficial to people to finish intelligent analysis of the monitoring data, improves the understanding of a computer to target vehicles, and further promotes the development of intelligent traffic.
The vehicle attributes mainly include vehicle types, colors, brands and the like, and the conventional vehicle identification problem process is to convert input original picture pixel values into manual features such as Scale-invariant feature (SIFT) features, Histogram of Oriented Gradients (HOG) features and the like, and then put the features obtained through conversion into a classifier (such as an SVM) for classification, so as to finally obtain an object identification result. The traditional algorithm is mainly classified based on manually defined features, and the identification accuracy is low. With the development of artificial intelligence technology, deep learning is popular with most companies with its innovative network structure and high accuracy, and has wide application in various fields such as security, finance, games, etc., but in practical application, the network model has high computational complexity, complex network structure and slow recognition speed. Moreover, the appearance of different brands of vehicles can be very similar when viewed from the same angle; and the shapes of vehicles of the same brand are greatly different from each other at different angles, so that the challenge is brought to the brand identification of the vehicles. Therefore, it is urgent to find a high-precision and high-speed recognition method.
Disclosure of Invention
In view of the problems in the prior art, the invention provides a lightweight vehicle attribute identification method based on metric learning, which mainly solves the problems of high complexity, low identification speed and low identification rate of a vehicle attribute identification network.
In order to achieve the above and other objects, the present invention adopts the following technical solutions.
A lightweight vehicle attribute identification method based on metric learning comprises the following steps:
training a lightweight convolutional neural network according to the multi-angle vehicle sample image to obtain a vehicle attribute identification model;
acquiring a vehicle image, and inputting the vehicle attribute identification model to acquire a vehicle attribute; wherein the vehicle attributes comprise model, color, brand information of the vehicle.
Optionally, the lightweight convolutional neural network comprises MobileNet V2.
Optionally, inputting the labeled sample image into a convolutional layer of the lightweight convolutional neural network for feature extraction, and acquiring a shared feature;
and filtering the shared features to obtain vehicle attribute features, and carrying out classification training on the vehicle attribute features to obtain the vehicle attribute recognition model.
Optionally, the shared features are filtered by setting a feature migration layer to connect the convolutional layer and the fully-connected layer of the lightweight convolutional neural network.
Optionally, a multi-angle loss function and a cross entropy loss function are respectively constructed to process the vehicle attribute features output by the full connection layer, and the multi-angle loss function and the cross entropy loss function are weighted and counted to obtain a comprehensive loss function for training the vehicle attribute identification model.
Optionally, according to the synthetic loss function, a back propagation algorithm is adopted to perform gradient back-pass layer by layer forwards, parameters of the lightweight convolutional neural network are trained, and the vehicle attribute identification model is obtained.
Optionally, selecting a sample from the sample image as a reference sample;
respectively selecting a similar sample belonging to the same class as the reference sample and a heterogeneous sample of different classes to form a triple;
forming a positive sample pair by the reference sample and the similar sample, and forming a negative sample pair by the reference sample and the heterogeneous sample;
creating a first loss function according to the similarity of the positive sample pair and the negative sample pair at the same angle; creating a second loss function according to the similarity of the positive sample pair and the negative sample pair at different angles; creating a third loss function according to the similarity between the negative sample pairs with the same angle and the positive sample pairs with different angles;
and acquiring the multi-angle loss function according to the first loss function, the second loss function and the third loss function.
And adjusting the threshold parameter of the distance interval between each sample pair aiming at the first loss function, the second loss function and the third loss function to obtain the optimal distance interval parameter.
Optionally, the lightweight convolutional neural network comprises an initial convolutional layer of 32 filters and 19 residual bottleneck layers.
Optionally, before the feature extraction, the sample image is normalized to obtain an image with a uniform size and gray scale range, and the image is input to the neural network.
Optionally, the metric learning-based lightweight vehicle attribute identification method according to claim 3, characterized in that ReLU6 is used as the nonlinear activation function of the convolutional layer, and the convolutional kernel size is 3X 3. As described above, the method for identifying lightweight vehicle attributes based on metric learning according to the present invention has the following advantages.
By integrating metric learning and the lightweight convolution neural network, the vehicle attribute identification accuracy can be effectively improved, and the identification speed is improved.
Drawings
Fig. 1 is a flowchart of a lightweight vehicle attribute identification method based on metric learning according to an embodiment of the present invention.
FIG. 2 is a flow chart of a lightweight vehicle attribute identification method based on metric learning according to another embodiment of the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
Referring to fig. 1, the present invention provides a lightweight vehicle attribute identification method based on metric learning, including steps S01-S02.
In step S01, training a lightweight convolutional neural network according to the multi-angle vehicle sample image, and obtaining a vehicle attribute identification model:
before the vehicle features are obtained, vehicle images of different angles can be collected and sorted to be used as vehicle sample images for model training. And (4) carrying out standardization processing on the vehicle sample image, and inputting the vehicle sample image into a neural network for feature extraction. The normalization process may be performed by cropping the sample image into image blocks of a predetermined size, such as 224 × 224 image blocks, and performing a gray-scale process on the cropped image. Taking an RGB image as an example, the average value of the gray scale of the corresponding color channel may be subtracted from each pixel in the RGB image.
In one embodiment, the neural network may use a MobileNet V2 network as a backbone network for feature extraction. MobileNet V2 is a lightweight network that uses mainly depth-wise convolution and point-wise convolution techniques. Wherein, the depth separable convolution is used for feature extraction, one convolution kernel of the depth separable convolution is responsible for one channel, and one channel is only convoluted by one convolution kernel. However, the number of feature maps after the depth separable convolution cannot be changed, and therefore, the increase and decrease of the feature maps are realized using the point convolution. In addition, the convolutional layer is used to extract vehicle features, and the pooling layer mainly down-samples the input information.
And a feature migration layer is arranged in front of the last convolution layer of the MobileNet V2 network, and the feature migration layer filters the vehicle features according to the vehicle attributes. A plurality of feature migration layers can be arranged, each feature migration layer is composed of convolution layers with convolution kernels of 1 x 1, the number of output channels is equal to the number of input channels, different positions are not mutually influenced, down sampling of an original feature diagram is avoided, and original feature dimensionality is kept. Wherein the vehicle attributes may include model, color, brand, etc.
After the vehicle features are filtered by the feature migration layer, the vehicle features are input into the identification branch corresponding to each vehicle attribute through the last convolutional layer of the MobileNet V2 network. Taking three attribute identification branches of vehicle type identification, color identification and brand identification as examples, a full connection layer can be respectively accessed behind a plurality of feature migration layers, and the number of full connection neurons is the number of categories of the vehicle attributes. And the full connection layers are connected to the characteristic migration layer corresponding to the attribute identification branches to obtain the characteristic vectors corresponding to the attribute identification branches.
In one embodiment, a lightweight convolutional neural network MobileNet V2 is used as a backbone network to extract the vehicle image features; the lightweight convolutional neural network consists of an initial full convolutional layer of 32 filters, 19 residual bottleneck layers. And each layer of feature map is connected to the local area of the previous layer of feature map through a convolution kernel, and the feature map of each layer is obtained through convolution operation and then through a weighting and activating function. ReLU6 is used as the non-linear activation function because it is robust when using low precision calculations. The convolution kernel size used is 3x3 and dropout and batch normalization are used during training.
In one embodiment, the deep neural network input that can employ the triplet structure is a triplet, which consists of three samples: one is to randomly select a sample from the training samples as a reference sample, then randomly select a sample belonging to the same class as the reference sample and a sample of a different class, and the two samples respectively form a similar sample and a heterogeneous sample of the reference sample. Three samples form a triplet and the entire network is then trained by the loss function.
The appearance of different vehicles may be very similar when viewed from the same angle; the shape of the same vehicle varies greatly from one angle to another. Therefore, for the three attribute recognition tasks of vehicle type recognition, color recognition and brand recognition, the multi-angle triplet comprehensive loss and the cross entropy loss are respectively followed after each branch subnetwork.
In an embodiment, creating the multi-angle triplet comprehensive loss through the spatial metric relationship of the homogeneous sample and the heterogeneous sample; the spatial metric can be calculated by Euclidean distance, cosine distance and the like.
Taking the Euclidean distance calculation similarity as an example, defining s as a similar angle, d as different angles, | ·| survival2Expression of Euclidean distance, Ps +A pair of positive samples representing the same angle,representing pairs of positive samples, P, at different angless -The pair of negative examples representing the same angle,negative examples representing different anglesCarrying out pairing; f. ofsAnd fdRepresenting feature space values for the same angle and different angles, respectively, β is a set threshold, which here may be expressed as a distance separation between sample pairs.
Specifically, the multi-angle triplet function consists of three losses:
Ltriplet=Ls+Ld+Ljoint
wherein L issRepresenting a first loss function, L, created at the same angledRepresenting a second loss function, L, created at a different anglejointRepresenting a third loss function created under the same angle and different angle conditions.
Definition Ds(P)=||fs(xi)-fs(xj)||2In particular, LsThe expression of (A) is as follows:
Ls=max{Ds(Ps +)-Ds(Ps -)+β,0}
by the above constraint, D will be causeds(Ps -) Greater than Ds(Ps +) I.e. heterogeneous samples at the same angle are far from each other and homogeneous samples are close to each other, and let D bes(Ps -) And Ds(Ps +) There is a minimum spacing β between the minimum spacing can be adjusted depending on the application.
Definition Dd(P)=||fd(xi)-fd(xj)||2In particular, LdThe expression of (A) is as follows:
Ld=max{Dd(Pd +)-Dd(Pd -)+β,0}
by the above constraint, makeIs greater thanI.e. differentThe heterogeneous samples of the angle are far away from each other, the homogeneous samples are close to each other, andandwith a minimum spacing β therebetween.
In the prior art, LsAnd LdThe loss is calculated in each independent angle feature space, and the mutual connection of the multi-angle feature spaces is omitted. Therefore, the loss is further constrained by a multi-angle loss calculation, specifically a third loss function LjointIs expressed as follows:
Ljoint=max{Dd(Pd +)-Ds(Ps -)+β,0}
by the above constraint, makeIs greater thanI.e. different kinds of samples at the same angle are far away from each other, and the same kinds of samples at different angles are close to each other, and letAnda minimum spacing β therebetween;
in summary, by LsAnd LdThe two triplets are loss-constrained, so that similar samples are close to each other and heterogeneous samples are far from each other in the feature space of respective angles. At the same time, LjointIn the feature spaces of different angles, the same-class samples are close to each other, and the different-class samples are far away from each other. And (4) performing statistical analysis on the three loss functions to create a multi-angle triplet comprehensive loss function.
Specifically, for three attribute identification tasks of vehicle type identification, color identification and brand identification, a softmax cross entropy is followed after the last full connection layer as a statistical loss function. Assuming that z is the probability vector of the input computational loss and f (z) is the output of softmax, then:
assuming y is the true class of z corresponding samples, the statistical loss function is defined as:
Lsoftmax=-logf(zy)
and fusing the multi-angle triplet comprehensive loss and the cross entropy loss as a loss function of vehicle attribute identification, setting the weight of the multi-angle triplet comprehensive loss function, and performing weighted statistics through the multi-angle triplet comprehensive loss function and the statistical loss function to obtain the loss function of the vehicle attribute classification model.
The penalty function for the vehicle attribute classification model may be expressed as:
L=αLtriplet+(1-α)Lsoftmax
wherein α is used for controlling the weight parameters of the two loss functions respectively, and the larger α is, the larger the proportion of the multi-angle triplet comprehensive loss function is.
And simultaneously using the multi-angle triplet comprehensive loss and the cross entropy loss in a model training process, so that each sub-network in the neural network learns corresponding independent characteristics, and the final fusion characteristics are richer. And performing test identification only through the cross entropy loss in a test stage.
In an embodiment, a random gradient descent method may be used to derive a loss function of the vehicle attribute classification model to obtain a network parameter gradient corresponding to the deep neural network, and then the deep neural network parameter is updated through the network parameter gradient, and the parameter is adjusted until the deep neural network converges to a stable state to obtain a final vehicle attribute identification model.
In step S02, a vehicle image is acquired, and a vehicle attribute identification model is input to acquire a vehicle attribute;
and inputting the vehicle images of a plurality of angles to be tested into the trained vehicle attribute recognition model, recognizing the characteristics corresponding to the vehicle attributes, and outputting the vehicle attributes including information of the vehicle type, the color, the brand and the like.
Referring to fig. 2, specifically, the method may include the following steps:
inputting a vehicle image 001; performing image preprocessing 002 to obtain a vehicle image with standardized size and gray scale;
extracting features of a standardized vehicle image 003 based on Mobilene V2 by taking Mobilene V2 as a backbone network;
multi-task attribute identification 004 based on a multi-angle loss function and a cross entropy loss function, and respectively identifying each vehicle attribute in the vehicle image through weight statistics of two loss functions;
vehicle attributes including model, color, brand, etc. are output 005.
In conclusion, the lightweight vehicle attribute identification method based on metric learning fully considers the relevance of the image features of the vehicles at different angles, and can effectively improve the identification accuracy through the training learning of two loss functions; the light weight neural network is adopted, so that the calculation amount in the identification process is reduced, and the identification efficiency is improved; meanwhile, the vehicle attribute recognition method is used for recognizing various vehicle attributes, and a feature sharing and feature transferring mode is adopted, so that calculation can be effectively saved, and the recognition efficiency is improved. Therefore, the invention effectively overcomes various defects in the prior art and has high industrial utilization value.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
Claims (10)
1. A lightweight vehicle attribute identification method based on metric learning is characterized by comprising the following steps:
training a lightweight convolutional neural network according to the multi-angle vehicle sample image to obtain a vehicle attribute identification model;
acquiring a vehicle image, and inputting the vehicle attribute identification model to acquire a vehicle attribute; wherein the vehicle attributes comprise model, color, brand information of the vehicle.
2. The metric learning-based lightweight vehicle attribute identification method of claim 1, wherein the lightweight convolutional neural network comprises MobileNet V2.
3. The lightweight vehicle attribute identification method based on metric learning of claim 1, characterized in that the labeled sample image is input into the convolutional layer of the lightweight convolutional neural network for feature extraction to obtain shared features;
and filtering the shared features to obtain vehicle attribute features, and carrying out classification training on the vehicle attribute features to obtain the vehicle attribute recognition model.
4. The metric learning-based lightweight vehicle attribute identification method according to claim 3, wherein the shared features are filtered by setting a feature migration layer to connect the convolutional layer and the fully-connected layer of the lightweight convolutional neural network.
5. The method for lightweight vehicle attribute recognition based on metric learning of claim 4, wherein a multi-angle loss function and a cross entropy loss function are respectively constructed to process the vehicle attribute features output by the full connection layer, and the multi-angle loss function and the cross entropy loss function are weighted and counted to obtain a comprehensive loss function for training the vehicle attribute recognition model.
6. The method according to claim 5, wherein a back propagation algorithm is used to perform gradient back-pass layer by layer forward according to the synthetic loss function, and parameters of the lightweight convolutional neural network are trained to obtain the vehicle attribute recognition model.
7. The metric learning-based lightweight vehicle attribute identification method of claim 5,
selecting a sample from the sample image as a reference sample;
respectively selecting a similar sample belonging to the same class as the reference sample and a heterogeneous sample of different classes to form a triple;
forming a positive sample pair by the reference sample and the similar sample, and forming a negative sample pair by the reference sample and the heterogeneous sample;
creating a first loss function according to the similarity of the positive sample pair and the negative sample pair at the same angle; creating a second loss function according to the similarity of the positive sample pair and the negative sample pair at different angles; creating a third loss function according to the similarity between the negative sample pairs with the same angle and the positive sample pairs with different angles;
and acquiring the multi-angle loss function according to the first loss function, the second loss function and the third loss function.
And adjusting the threshold parameter of the distance interval between each sample pair aiming at the first loss function, the second loss function and the third loss function to obtain the optimal distance interval parameter.
8. The metric learning-based lightweight vehicle attribute identification method according to claim 2,
the lightweight convolutional neural network includes an initial convolutional layer of 32 filters and 19 residual bottleneck layers.
9. The metric learning-based lightweight vehicle attribute identification method of claim 3,
and carrying out standardization processing on the sample image before the characteristic extraction is carried out, and obtaining an image input neural network with uniform size and gray scale range.
10. The metric learning-based lightweight vehicle attribute identification method of claim 3, characterized in that ReLU6 is employed as the nonlinear activation function of the convolutional layer, and the convolutional kernel size is 3X 3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911234612.6A CN110991349B (en) | 2019-12-05 | 2019-12-05 | Lightweight vehicle attribute identification method based on metric learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911234612.6A CN110991349B (en) | 2019-12-05 | 2019-12-05 | Lightweight vehicle attribute identification method based on metric learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110991349A true CN110991349A (en) | 2020-04-10 |
CN110991349B CN110991349B (en) | 2023-02-10 |
Family
ID=70090386
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911234612.6A Active CN110991349B (en) | 2019-12-05 | 2019-12-05 | Lightweight vehicle attribute identification method based on metric learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110991349B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111814584A (en) * | 2020-06-18 | 2020-10-23 | 北京交通大学 | Vehicle weight identification method under multi-view-angle environment based on multi-center measurement loss |
CN111931668A (en) * | 2020-08-14 | 2020-11-13 | 中国科学院重庆绿色智能技术研究院 | Target attribute identification method and system based on adaptive convolutional neural network |
CN111931768A (en) * | 2020-08-14 | 2020-11-13 | 中国科学院重庆绿色智能技术研究院 | Vehicle identification method and system capable of self-adapting to sample distribution |
CN112016490A (en) * | 2020-08-28 | 2020-12-01 | 中国科学院重庆绿色智能技术研究院 | Pedestrian attribute identification method based on generation countermeasure learning |
CN112329785A (en) * | 2020-11-25 | 2021-02-05 | Oppo广东移动通信有限公司 | Image management method, device, terminal and storage medium |
CN115131768A (en) * | 2022-04-29 | 2022-09-30 | 浙江大华技术股份有限公司 | Training method of vehicle attribute identification network, and vehicle attribute identification method and device |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140098989A1 (en) * | 2012-10-05 | 2014-04-10 | International Business Machines Corporation | Multi-cue object association |
CN105654066A (en) * | 2016-02-02 | 2016-06-08 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
CN106599869A (en) * | 2016-12-22 | 2017-04-26 | 安徽大学 | Vehicle attribute identification method based on multi-task convolutional neural network |
CN106682649A (en) * | 2017-01-24 | 2017-05-17 | 成都容豪电子信息科技有限公司 | Vehicle type recognition method based on deep learning |
CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
CN108492575A (en) * | 2018-04-11 | 2018-09-04 | 济南浪潮高新科技投资发展有限公司 | A kind of intelligent vehicle type identifier method |
CN108549926A (en) * | 2018-03-09 | 2018-09-18 | 中山大学 | A kind of deep neural network and training method for refining identification vehicle attribute |
CN109359684A (en) * | 2018-10-17 | 2019-02-19 | 苏州大学 | Fine granularity model recognizing method based on Weakly supervised positioning and subclass similarity measurement |
CN109815799A (en) * | 2018-12-18 | 2019-05-28 | 南京理工大学 | A kind of vehicle detecting algorithm of quickly taking photo by plane based on SSD |
CN110147709A (en) * | 2018-11-02 | 2019-08-20 | 腾讯科技(深圳)有限公司 | Training method, device, terminal and the storage medium of vehicle attribute model |
CN110309770A (en) * | 2019-06-28 | 2019-10-08 | 华侨大学 | A kind of vehicle discrimination method again based on the study of four-tuple loss metric |
CN110334572A (en) * | 2019-04-04 | 2019-10-15 | 南京航空航天大学 | The fine recognition methods of vehicle under a kind of multi-angle |
CN110399828A (en) * | 2019-07-23 | 2019-11-01 | 吉林大学 | A kind of vehicle recognition methods again of the depth convolutional neural networks based on multi-angle |
CN110414578A (en) * | 2019-07-16 | 2019-11-05 | 上海电机学院 | A kind of transfer learning method based on the multiple batches of training of dynamic and colour gamut conversion |
-
2019
- 2019-12-05 CN CN201911234612.6A patent/CN110991349B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140098989A1 (en) * | 2012-10-05 | 2014-04-10 | International Business Machines Corporation | Multi-cue object association |
CN105654066A (en) * | 2016-02-02 | 2016-06-08 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
CN106599869A (en) * | 2016-12-22 | 2017-04-26 | 安徽大学 | Vehicle attribute identification method based on multi-task convolutional neural network |
CN106682649A (en) * | 2017-01-24 | 2017-05-17 | 成都容豪电子信息科技有限公司 | Vehicle type recognition method based on deep learning |
CN107886073A (en) * | 2017-11-10 | 2018-04-06 | 重庆邮电大学 | A kind of more attribute recognition approaches of fine granularity vehicle based on convolutional neural networks |
CN108549926A (en) * | 2018-03-09 | 2018-09-18 | 中山大学 | A kind of deep neural network and training method for refining identification vehicle attribute |
CN108492575A (en) * | 2018-04-11 | 2018-09-04 | 济南浪潮高新科技投资发展有限公司 | A kind of intelligent vehicle type identifier method |
CN109359684A (en) * | 2018-10-17 | 2019-02-19 | 苏州大学 | Fine granularity model recognizing method based on Weakly supervised positioning and subclass similarity measurement |
CN110147709A (en) * | 2018-11-02 | 2019-08-20 | 腾讯科技(深圳)有限公司 | Training method, device, terminal and the storage medium of vehicle attribute model |
CN109815799A (en) * | 2018-12-18 | 2019-05-28 | 南京理工大学 | A kind of vehicle detecting algorithm of quickly taking photo by plane based on SSD |
CN110334572A (en) * | 2019-04-04 | 2019-10-15 | 南京航空航天大学 | The fine recognition methods of vehicle under a kind of multi-angle |
CN110309770A (en) * | 2019-06-28 | 2019-10-08 | 华侨大学 | A kind of vehicle discrimination method again based on the study of four-tuple loss metric |
CN110414578A (en) * | 2019-07-16 | 2019-11-05 | 上海电机学院 | A kind of transfer learning method based on the multiple batches of training of dynamic and colour gamut conversion |
CN110399828A (en) * | 2019-07-23 | 2019-11-01 | 吉林大学 | A kind of vehicle recognition methods again of the depth convolutional neural networks based on multi-angle |
Non-Patent Citations (4)
Title |
---|
CALVINPAEAN: "MobileNetV2 论文学习", 《HTTPS://BLOG.CSDN.NET/CALVINPAEAN/ARTICLE/DETAILS/83994206》 * |
MARK SANDLER 等: "MobileNetV2: Inverted Residuals and Linear Bottlenecks", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
林唯贤: "嵌入式设备高效卷积神经网络的电力设备检测", 《计算机系统应用》 * |
王耀玮 等: "基于多任务卷积神经网络的车辆多属性识别", 《计算机工程与应用》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111814584A (en) * | 2020-06-18 | 2020-10-23 | 北京交通大学 | Vehicle weight identification method under multi-view-angle environment based on multi-center measurement loss |
CN111814584B (en) * | 2020-06-18 | 2024-03-12 | 北京交通大学 | Vehicle re-identification method based on multi-center measurement loss under multi-view environment |
CN111931668A (en) * | 2020-08-14 | 2020-11-13 | 中国科学院重庆绿色智能技术研究院 | Target attribute identification method and system based on adaptive convolutional neural network |
CN111931768A (en) * | 2020-08-14 | 2020-11-13 | 中国科学院重庆绿色智能技术研究院 | Vehicle identification method and system capable of self-adapting to sample distribution |
CN112016490A (en) * | 2020-08-28 | 2020-12-01 | 中国科学院重庆绿色智能技术研究院 | Pedestrian attribute identification method based on generation countermeasure learning |
CN112329785A (en) * | 2020-11-25 | 2021-02-05 | Oppo广东移动通信有限公司 | Image management method, device, terminal and storage medium |
CN115131768A (en) * | 2022-04-29 | 2022-09-30 | 浙江大华技术股份有限公司 | Training method of vehicle attribute identification network, and vehicle attribute identification method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110991349B (en) | 2023-02-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110991349B (en) | Lightweight vehicle attribute identification method based on metric learning | |
CN108182441B (en) | Parallel multichannel convolutional neural network, construction method and image feature extraction method | |
CN107066559B (en) | Three-dimensional model retrieval method based on deep learning | |
CN108717524B (en) | Gesture recognition system based on double-camera mobile phone and artificial intelligence system | |
CN106919920B (en) | Scene recognition method based on convolution characteristics and space vision bag-of-words model | |
CN112633350B (en) | Multi-scale point cloud classification implementation method based on graph convolution | |
CN111476266B (en) | Non-equilibrium type leukocyte classification method based on transfer learning | |
CN110008842A (en) | A kind of pedestrian's recognition methods again for more losing Fusion Model based on depth | |
CN110175615B (en) | Model training method, domain-adaptive visual position identification method and device | |
CN112580590A (en) | Finger vein identification method based on multi-semantic feature fusion network | |
CN110728179A (en) | Pig face identification method adopting multi-path convolutional neural network | |
CN112801015B (en) | Multi-mode face recognition method based on attention mechanism | |
CN110532946B (en) | Method for identifying axle type of green-traffic vehicle based on convolutional neural network | |
CN110222718B (en) | Image processing method and device | |
CN110728694B (en) | Long-time visual target tracking method based on continuous learning | |
CN113780132B (en) | Lane line detection method based on convolutional neural network | |
CN110032925A (en) | A kind of images of gestures segmentation and recognition methods based on improvement capsule network and algorithm | |
CN113128308B (en) | Pedestrian detection method, device, equipment and medium in port scene | |
CN109165698A (en) | A kind of image classification recognition methods and its storage medium towards wisdom traffic | |
CN113420794B (en) | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning | |
CN110046544A (en) | Digital gesture identification method based on convolutional neural networks | |
CN114463812B (en) | Low-resolution face recognition method based on double-channel multi-branch fusion feature distillation | |
CN115049814B (en) | Intelligent eye protection lamp adjusting method adopting neural network model | |
CN112633257A (en) | Potato disease identification method based on improved convolutional neural network | |
CN110211127A (en) | Image partition method based on bicoherence network |
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 |