CN110176143A - A kind of highway traffic congestion detection method based on deep learning algorithm - Google Patents
A kind of highway traffic congestion detection method based on deep learning algorithm Download PDFInfo
- Publication number
- CN110176143A CN110176143A CN201910497362.9A CN201910497362A CN110176143A CN 110176143 A CN110176143 A CN 110176143A CN 201910497362 A CN201910497362 A CN 201910497362A CN 110176143 A CN110176143 A CN 110176143A
- Authority
- CN
- China
- Prior art keywords
- picture
- congestion
- alexnet
- traffic congestion
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- 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
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0133—Traffic data processing for classifying traffic situation
-
- 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
- G08G1/04—Detecting movement of traffic to be counted or controlled using optical or ultrasonic detectors
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of highway traffic congestion detection method based on deep learning algorithm, classifier based on convolutional neural networks (CNN), the full articulamentum in AlexNet and VGGNet is replaced using support vector machines (SVM), two classification, i.e. congestion and not congestion are carried out to traffic status of express way.The present invention compares the detection accuracy and calculating speed of two kinds of algorithms, determines that the effect of the AlexNet+SVM with transfer learning is optimal.
Description
Technical field
The invention belongs to highway detection technique field, in particular to a kind of highway based on deep learning algorithm
Traffic jam detection method.
Background technique
With traffic jam issue become increasingly conspicuous and the continuous development of intelligent transportation system (ITS), highway hand over
The real-time detection of logical congestion becomes the research hotspot handed over now.Highway congestion detection to traffic control, paths chosen and
Line efficiency is lifted out to be of great significance.Due to the complexity of current traffic environment, most of detection methods be difficult according to illumination,
Situations such as weather, carries out automatic identification and report predicted traffic jam situation, therefore disturbed condition how to be overcome accurately to judge traffic shape
State is the key that solve traffic jam issue.
Two kinds can be substantially divided into for the detection method of traffic congestion at present.One is the congestion inspections based on car data
It surveys, car data can mainly use vehicle checker and roadside sensors pair for the GPS data of taxi, bus etc., which
Car data is acquired, and carries out congestion differentiation using collected data, but due on China's highway, vehicle checker
Installation is than sparse and some abilities of having lost the job, therefore the data got, there are abnormal and incomplete, this is to real-time
Differentiate that traffic congestion state brings very big problem.If the such as document [1] Zhang Yaru, Zhao Haitao, Liu Nanjie are based on GPS data
Traffic congestion detects [J] computer technology and development, 2017,27 (1): 139-142, document [2] Li Yong are based on taxi GPS
The urban traffic blocking identification of data and the Harbin association analysis [D]: Harbin Institute of Technology, 2016.Document [1], [2]
It is all to be detected using GPS data to traffic congestion.Document [1] cleans raw GPS data and is repaired first, that is, counts
Secondly Data preprocess has carried out clustering to GPS data using K-means algorithm, has realized the survey region under different clusters
It divides, finally uses the Traffic State Detection Method that sample size carries out random sampling in local cluster and traffic behavior is carried out
Differentiate.Document [2] is handled by pretreatment GPS data from taxi, trackization, divides road network and period, split track, calculate rail
Mark speed and the average speed for calculating the daily each grid of day part, identify the traffic congestion region of daily day part.It should
Method can differentiate traffic congestion, but huge calculation amount causes it effective right with cumbersome data prediction
Traffic congestion carries out real time discriminating, and the loss of data has very big otherness with the abnormal result that even more will lead to, and precision is difficult to
It is promoted.
Second is the congestion detection based on image, and this method mainly utilizes freeway surveillance and control camera to obtain video
Image handles image, carries out congestion differentiation with neural network algorithm.This method speed is fast, obtains data simplicity, is
The research hotspot that congestion differentiates.Road vehicle jamming analysis model [J] such as document [3] Zhang Jiachen, Chen Qingkui based on YOLO
Computer application, traffic congestion detection algorithm [J] of 2019,39 (1), document [4] Yuan Bin, Zhang Yong based on analyzing image texture
Shanghai Ship & Shipping Research Institute journal, 2015,38 (4): 77-81. document [3] is based on YOLOv3 algorithm of target detection, knot
Close the corresponding eigenvalue matrix of picture, the result for making the difference by the eigenmatrix between consecutive frame and difference being summed item by item
It is compared to judge that present road is to be in congestion status or normal pass state with preset value, it can be simultaneously to one
Three lanes of road carry out statistic analysis, can reach 80% to the judging nicety rate in single lane.Document [4] proposes one
Road traffic congestion detection algorithm of the kind based on analyzing image texture, the algorithm is using image block as basic processing unit, by dividing
The spatially and temporally variation for analysing image texture in block calculates space and the time occupancy of road, and both comprehensive variation
The automatically and rapidly detection of characteristic realization road traffic congestion.Due to monitoring camera obtain picture quality it is irregular and
Complicated and changeable, the problems such as that there are precision mostly is not high for congestion based on image detection, and false detection rate is higher of traffic scene.
The processing of technical data disclosed in above-mentioned document is cumbersome, calculates complexity, and timeliness is low, detection accuracy deficiency etc., because
This is difficult to apply in a practical situation.
Summary of the invention
The purpose of the present invention is to provide a kind of highway traffic congestion detection method based on deep learning algorithm, with
It solves the above problems.
To achieve the above object, the invention adopts the following technical scheme:
A kind of highway traffic congestion detection method based on deep learning algorithm, comprising the following steps:
Step 1, video data is obtained from highway monitoring system, by video data interception at picture, to picture into
Row pretreatment;
Step 2, the traffic congestion situation reported according to highway work personnel determines congestion picture and non-congestion figure
Piece establishes image classification data library;
Step 3, using deep learning sorting algorithm AlexNet, the good AlexNet network structure of pre-training is chosen as this
The foundation structure of disaggregated model;
Step 4, network structure is improved, uses the full articulamentum that SVM substitution AlexNet is last;
Step 5, verifying and small parameter perturbations are trained to improved network model, using picture database 80% as
Training set, it is remaining to be used as test set;Determine the optimal value of learning rate, momentum, descending factors and decline cycle;
Step 6, trained Model Weight is saved, the picture database data of residue 20% is tested;
The number of iterations and penalty values are drawn, the relation curve of the number of iterations and training precision judges network effect;
Step 7, new data set is tested, judges traffic congestion state.
Further, in step 1, every section of the video data of acquisition when, is 10 minutes a length of, and video data is intercepted as picture,
Time interval is 1s or 2s, pre-processes to picture, rejects unintelligible picture, do not take highway picture and camera shooting
Machine repairs picture, and all picture sizes are changed to 227*227.
Further, 30000 congestion images and 20000 non-congestion images are chosen in step 2, by congestion picture with it is non-
Congestion picture deposits in file 0 and file 1 respectively, establishes traffic congestion picture database, and 0 is does not block up, and 1 is stifled.
Further, deep learning sorting algorithm AlexNet, arrange in pairs or groups convolutional neural networks framework CNN, using 11 ×
The convolution process of 11 × 3 convolution kernel, step-length 4, CNN can be expressed as formula:
F [x, y] is a part of original image, and g [x, y] is CNN convolution kernel, and in the present invention, x=11, y=11, * is
Convolution process, n1,n2For the pixel in convolution region, x is the dot product of each pixel;
The input of AlexNet is 227 × 227 × 3 image, and image is carried out by 5 layers of CNN structure and 3 layers of full articulamentum
Training after all inputs are summed with weight, reduces calculation amount using ReLU activation primitive, weight sum formula is as follows:
yi=f (∑iωixi+ b) formula (2)
Wherein, ωiFor the weight of convolution kernel, xiFor the value of each lattice of image array, b is bias.
ReLU activation primitive formula is as follows:
F (x)=max (0, x) formula (3)
Further, in step 5, verifying is trained to network using computer, determines that each parameter situation is as follows: study
Rate (learning rate) is 0.0001, and momentum (momentum) is 0.9, and descending factors (drop factor) are 0.05, under
Dropping period (drop period) is 5.
Compared with prior art, the present invention has followingTechnical effect:
The present invention uses transfer learning, improves to traditional AlexNet network structure, substitutes AlexNet using SVM
Full articulamentum, relatively traditional congestion detection method and unmodified algorithm not only ensure that the precision of detection, but also greatly reduce
Calculating speed specifically has the following advantages,
First: the selection at interception picture interval.Present invention determine that the time interval of video intercepting picture is 1s, both ensure that
The quantity of picture in turn ensures the otherness of picture;
Second: congestion pictures being determined according to the jam situation that staff reports, are classified just when ensure that picture training
Really;
Third: congestion picture and non-congestion picture are respectively put into file 1 and file 0, keep whole process simply easy
Operation, ensure that the quality and operability of database;
4th: choosing the good network structure of pre-training as infrastructure network of the invention, ensure that training precision
While reduce the data training time;
5th: using the full articulamentum of SVM substitution AlexNet, improving detection accuracy.
Using the AlexNet+SVM model, is verified by training and carry out small parameter perturbations, it is ensured that the conjunction of its parameter designing
Rationality and correctness.
By to AlexNet+SVM, VGGNet+SVM, with AlexNet+SVM, VGGNet+SVM mould of transfer learning
Type is trained test respectively, and contrast test precision and training time find optimal algorithm and model, it is ensured that present invention detection
Effect is optimal.
By visualizing to each convolutional layer, guarantee the accuracy of the model extraction feature, it is ensured that model key feature
It extracts without relatively large deviation.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Fig. 2 is established highway congestion picture database.
Fig. 3 is the convolution process of 11 × 11 × 3 convolution kernels.
Fig. 4 is the network structure of AlexNet.
Fig. 5 is the present invention to the improved network structure of AlexNet.
Fig. 6 is the relation curve of the number of iterations and training loss, training precision in network training process.
Specific embodiment
A specific embodiment of the invention is described in further detail now in conjunction with Figure of description.
A kind of highway congestion detection method based on deep learning algorithm proposed by the present invention, first from highway
Monitoring camera obtains video data, and video data is processed into picture, is pre-processed to picture;Secondly by transfer learning
The good AlexNet network of pre-training is chosen as infrastructure network of the invention, the full articulamentum of AlexNet is replaced with
SVM;Verifying finally is trained to network, determines parameter value, congestion two is carried out to picture and is classified.
As shown in Figure 1, method flow are as follows:
Step 1: the present invention obtains video data from highway monitoring system, and video data is throughout highway road
Duan Suoyou video camera, every section of video length about 10 minutes;Application program intercepts video data for picture, time interval 1s,
Guarantee the quantity and quality of picture;Picture is pre-processed, rejecting wrong picture, (unintelligible picture does not take high speed public affairs
Road picture and video camera repair picture);All picture sizes are changed to 227*227, the i.e. required size of model.
Step 2: traffic congestion boundary has strong subjective consciousness, and the present invention reports situation determination to gather around according to staff
Stifled and non-congestion pictures are respectively put into file 1 and file 0, i.e., 0 is not congestion, and 1 is congestion;It is final to choose 30000
Congestion picture is opened, 20000 non-congestion pictures establish traffic congestion picture database, such as Fig. 2.
Step 3: the present invention uses deep learning sorting algorithm AlexNet, convolutional neural networks framework of having arranged in pairs or groups
(CNN), ReLU, Dropout and LRN are applied in CNN for the first time, while AlexNet can be used GPU progress operation and add
Speed;AlexNet uses activation primitive of the ReLU as CNN, and demonstrating the effect in deeper network has been more than Sigmoid
Function solves the problems, such as the gradient disperse in deeper network;AlexNet, using the maximum pond of overlapping, is avoided average in CNN
The blurring effect in pond, and in AlexNet step-length ratio Chi Huahe size it is small, have overlapping between the output of such pond layer
And covering, improve the rich of feature;It also proposed LRN layers, to the activity creation competition mechanism of local neuron, so that
Wherein the biggish value of response ratio becomes relatively bigger, and other is inhibited to feed back lesser neuron, enhances the extensive energy of model
Power.The present invention uses 11 × 11 × 3 convolution kernel, step-length 4, and Fig. 3 is CNN convolution process, and the convolution process of CNN can be with table
It is shown as following formula:
F [x, y] is a part of original image, and g [x, y] is CNN convolution kernel, and in the present invention, x=11, y=11, * is
Convolution process, n1,n2For the pixel in convolution region, x is the dot product of each pixel.
The input of AlexNet is 227 × 227 × 3 image, and image is carried out by 5 layers of CNN structure and 3 layers of full articulamentum
Training after all inputs are summed with weight, reduces calculation amount using ReLU activation primitive.Weight sum formula is as follows:
yi=f (∑iωixi+b)
Formula (2)
Wherein, ωiFor the weight of convolution kernel, xiFor the value of each lattice of image array, b is bias.
ReLU activation primitive formula is as follows:
F (x)=max (0, x)
Formula (3)
As shown in figure 4, the network structure of AlexNet are as follows:
Step 4: new network needs millions of pictures to be trained, and process is time-consuming and need not.The present invention uses
The AlexNet network structure that pre-training is crossed has been provided with some picture basis features, each node as basic network structure
Have some weights;By last three full articulamentums of SVM replacement AlexNet, such as Fig. 5.
Step 5: verifying being trained to improved network model, using picture database 80% as training set, is remained
Under be used as test set;Using Intel i7CPU@4.2GHz、GPU1080 computer is trained verifying to network.It is special
Not, determine that each parameter situation is as follows: learning rate (learning rate) is 0.0001, and momentum (momentum) is 0.9, under
Dropping the factor (drop factor) is 0.05, and decline cycle (drop period) is 5;Training penalty values, training precision and iteration
Frequency curve relationship such as Fig. 6.
Step 6: trained Model Weight being saved, the picture database data of residue 20% are tested.
The present invention has higher congestion detection accuracy and quick training speed to verify the model, will be with transfer learning
AlexNet+SVM is respectively with the VGGNet+SVM of utilization transfer learning, without using AlexNet+SVM, VGGNet of transfer learning
+ SVM compares verifying, discovery with the AlexNet+SVM of transfer learning model measurement precision highest, can reach 90% and
Training speed is only 30 minutes;Concrete outcome is as shown in table 1:
Each network model Contrast on effect table of table 1
The present invention is further tested picture, and measuring accuracy can maintain 90% or so, and with the number of iterations
Increase precision be also continuously increased;And 90% can be reached to the verification and measurement ratio under the disturbed conditions such as different illumination, weather, this hair
Bright improved network structure can be applied in current highway monitoring system, realize the detection to traffic congestion and report
It accuses.
Claims (5)
1. a kind of highway traffic congestion detection method based on deep learning algorithm, which comprises the following steps:
Step 1, video data is obtained from highway monitoring system, by video data interception at picture, picture is carried out pre-
Processing;
Step 2, the traffic congestion situation reported according to highway work personnel determines congestion picture and non-congestion picture, builds
Vertical image classification data library;
Step 3, using deep learning sorting algorithm AlexNet, the good AlexNet network structure of pre-training is chosen as this classification
The foundation structure of model;
Step 4, network structure is improved, uses the full articulamentum that SVM substitution AlexNet is last;
Step 5, verifying and small parameter perturbations are trained to improved network model, using picture database 80% as training
Collection, it is remaining to be used as test set;Determine the optimal value of learning rate, momentum, descending factors and decline cycle;
Step 6, trained Model Weight is saved, the picture database data of residue 20% is tested;It draws
The relation curve of the number of iterations and penalty values, the number of iterations and training precision, judges network effect;
Step 7, new data set is tested, judges traffic congestion state.
2. a kind of highway traffic congestion detection method based on deep learning algorithm according to claim 1, special
Sign is, in step 1, every section of the video data of acquisition when is 10 minutes a length of, and video data is intercepted as picture, and time interval is
1s or 2s, pre-processes picture, rejects unintelligible picture, does not take highway picture and video camera service action drawing
All picture sizes are changed to 227*227 by piece.
3. a kind of highway traffic congestion detection method based on deep learning algorithm according to claim 1, special
Sign is, 30000 congestion images and 20000 non-congestion images are chosen in step 2, by congestion picture and non-congestion picture point
File 0 and file 1 are not deposited in, establishes traffic congestion picture database, and 0 is does not block up, and 1 is stifled.
4. a kind of highway traffic congestion detection method based on deep learning algorithm according to claim 1, special
Sign is that deep learning sorting algorithm AlexNet, arrange in pairs or groups convolutional neural networks framework CNN, using 11 × 11 × 3 convolution
The convolution process of core, step-length 4, CNN can be expressed as formula:
F [x, y] is a part of original image, and g [x, y] is CNN convolution kernel, in the present invention, x=11, y=11,For convolution
Process, n1,n2For the pixel in convolution region, x is the dot product of each pixel;
The input of AlexNet is 227 × 227 × 3 image, and image is trained by 5 layers of CNN structure and 3 layers of full articulamentum,
After all inputs are summed with weight, calculation amount is reduced using ReLU activation primitive, weight sum formula is as follows:
yi=f (∑iωixi+ b) formula (2)
Wherein, ωiFor the weight of convolution kernel, xiFor the value of each lattice of image array, b is bias;
ReLU activation primitive formula is as follows:
F (x)=max (0, x) formula (3).
5. a kind of highway traffic congestion detection method based on deep learning algorithm according to claim 1, special
Sign is, in step 5, is trained verifying to network using computer, determines that each parameter situation is as follows: learning rate learning
Rate is 0.0001, momentum momentum be 0.9, descending factors drop factor be 0.05, decline cycle drop period
It is 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910497362.9A CN110176143A (en) | 2019-06-10 | 2019-06-10 | A kind of highway traffic congestion detection method based on deep learning algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910497362.9A CN110176143A (en) | 2019-06-10 | 2019-06-10 | A kind of highway traffic congestion detection method based on deep learning algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110176143A true CN110176143A (en) | 2019-08-27 |
Family
ID=67698045
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910497362.9A Pending CN110176143A (en) | 2019-06-10 | 2019-06-10 | A kind of highway traffic congestion detection method based on deep learning algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110176143A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110517497A (en) * | 2019-09-05 | 2019-11-29 | 中国科学院长春光学精密机械与物理研究所 | A kind of road traffic classification method, device, equipment, medium |
CN110533098A (en) * | 2019-08-28 | 2019-12-03 | 长安大学 | A method of identifying that the green compartment that is open to traffic loads type based on convolutional neural networks |
CN111368931A (en) * | 2020-03-09 | 2020-07-03 | 第四范式(北京)技术有限公司 | Method and device for training image classification model, computer device and storage medium |
CN113033383A (en) * | 2021-03-23 | 2021-06-25 | 山东大学 | Permeable pavement blocking nondestructive detection method and system based on deep learning |
CN113076893A (en) * | 2021-04-09 | 2021-07-06 | 太原理工大学 | Highway drain pipe blocking situation sensing method based on deep learning |
CN117292552A (en) * | 2023-11-27 | 2023-12-26 | 深圳市诚识科技有限公司 | High-speed road condition analysis system and method based on machine vision |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018030772A1 (en) * | 2016-08-10 | 2018-02-15 | 중앙대학교 산학협력단 | Responsive traffic signal control method and apparatus therefor |
CN108510739A (en) * | 2018-04-28 | 2018-09-07 | 重庆交通大学 | A kind of road traffic state recognition methods, system and storage medium |
-
2019
- 2019-06-10 CN CN201910497362.9A patent/CN110176143A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018030772A1 (en) * | 2016-08-10 | 2018-02-15 | 중앙대학교 산학협력단 | Responsive traffic signal control method and apparatus therefor |
CN108510739A (en) * | 2018-04-28 | 2018-09-07 | 重庆交通大学 | A kind of road traffic state recognition methods, system and storage medium |
Non-Patent Citations (4)
Title |
---|
倪铮,文韬: "一种基于CNN和RNN深度神经网络的天气预测模型—以北京地区雷暴的 6小时临近预报为何", 《数值计算与计算机应用》 * |
周飞燕,金林鹏,董军: "卷积神经网络研究综述", 《计算机学报》 * |
宋佳蓉,杨忠,张天翼,韩家明,朱家远: "基于卷积神经网络和多类SVM 的交通标志识别", 《应用科技》 * |
纪宇: "基于AlexNet模型的高速公路拥堵状态识别", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533098A (en) * | 2019-08-28 | 2019-12-03 | 长安大学 | A method of identifying that the green compartment that is open to traffic loads type based on convolutional neural networks |
CN110533098B (en) * | 2019-08-28 | 2022-03-29 | 长安大学 | Method for identifying loading type of green traffic vehicle compartment based on convolutional neural network |
CN110517497A (en) * | 2019-09-05 | 2019-11-29 | 中国科学院长春光学精密机械与物理研究所 | A kind of road traffic classification method, device, equipment, medium |
CN111368931A (en) * | 2020-03-09 | 2020-07-03 | 第四范式(北京)技术有限公司 | Method and device for training image classification model, computer device and storage medium |
CN111368931B (en) * | 2020-03-09 | 2023-11-17 | 第四范式(北京)技术有限公司 | Method for determining learning rate of image classification model |
CN113033383A (en) * | 2021-03-23 | 2021-06-25 | 山东大学 | Permeable pavement blocking nondestructive detection method and system based on deep learning |
CN113076893A (en) * | 2021-04-09 | 2021-07-06 | 太原理工大学 | Highway drain pipe blocking situation sensing method based on deep learning |
CN117292552A (en) * | 2023-11-27 | 2023-12-26 | 深圳市诚识科技有限公司 | High-speed road condition analysis system and method based on machine vision |
CN117292552B (en) * | 2023-11-27 | 2024-02-09 | 深圳市诚识科技有限公司 | High-speed road condition analysis system and method based on machine vision |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110176143A (en) | A kind of highway traffic congestion detection method based on deep learning algorithm | |
CN110310261B (en) | Contact net dropper defect detection model training method and defect detection method | |
CN106504233B (en) | Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN | |
CN110390691A (en) | A kind of ore scale measurement method and application system based on deep learning | |
CN111275688A (en) | Small target detection method based on context feature fusion screening of attention mechanism | |
CN110135296A (en) | Airfield runway FOD detection method based on convolutional neural networks | |
Li et al. | Sewer pipe defect detection via deep learning with local and global feature fusion | |
CN110189317A (en) | A kind of road image intelligent acquisition and recognition methods based on deep learning | |
CN112233073A (en) | Real-time detection method for infrared thermal imaging abnormity of power transformation equipment | |
CN110070530A (en) | A kind of powerline ice-covering detection method based on deep neural network | |
CN111444939A (en) | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field | |
CN109934805A (en) | A kind of water pollution detection method based on low-light (level) image and neural network | |
CN105138982A (en) | Crowd abnormity detection and evaluation method based on multi-characteristic cluster and classification | |
CN108596203A (en) | Optimization method of the pond layer in parallel to pantograph carbon slide surface abrasion detection model | |
CN116310785B (en) | Unmanned aerial vehicle image pavement disease detection method based on YOLO v4 | |
CN110222593A (en) | A kind of vehicle real-time detection method based on small-scale neural network | |
CN103679214B (en) | Vehicle checking method based on online Class area estimation and multiple features Decision fusion | |
CN105261148A (en) | Trample event early warning evacuation method based on skynet monitoring system | |
CN106650823A (en) | Probability extreme learning machine integration-based foam nickel surface defect classification method | |
CN109886937A (en) | Defects of insulator detection method based on super-pixel segmentation image recognition | |
CN109086803A (en) | A kind of haze visibility detection system and method based on deep learning and the personalized factor | |
CN108549901A (en) | A kind of iteratively faster object detection method based on deep learning | |
CN110059076A (en) | A kind of Mishap Database semi-automation method for building up of power transmission and transformation line equipment | |
CN103489012A (en) | Crowd density detecting method and system based on support vector machine | |
CN111191610A (en) | People flow detection and processing method in video monitoring |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190827 |