CN112528934A - Improved YOLOv3 traffic sign detection method based on multi-scale feature layer - Google Patents

Improved YOLOv3 traffic sign detection method based on multi-scale feature layer Download PDF

Info

Publication number
CN112528934A
CN112528934A CN202011527413.7A CN202011527413A CN112528934A CN 112528934 A CN112528934 A CN 112528934A CN 202011527413 A CN202011527413 A CN 202011527413A CN 112528934 A CN112528934 A CN 112528934A
Authority
CN
China
Prior art keywords
network
traffic sign
training
improved
yolov3
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
Application number
CN202011527413.7A
Other languages
Chinese (zh)
Inventor
李国强
付乐
孙英家
方奇
王天雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yanshan University
Original Assignee
Yanshan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yanshan University filed Critical Yanshan University
Priority to CN202011527413.7A priority Critical patent/CN112528934A/en
Publication of CN112528934A publication Critical patent/CN112528934A/en
Pending legal-status Critical Current

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
    • G06V20/56—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
    • G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • 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
    • 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

Abstract

The invention discloses a traffic sign detection method of improved YOLOv3 based on a multi-scale feature layer, which comprises the following steps: step 1, preparing a traffic sign data set and performing data enhancement; step 2, building a YOLOv3 improved network model, improving a backbone network, replacing the original Darknet53 with a Densenet network, and optimizing the Densenet network; step 3, training the improved network model of YOLOv 3; and 4, detecting the traffic sign by using the training optimal model. The invention has the advantages of balanced sample types, strong detection capability for different target scales and capability of detecting smaller traffic signs.

Description

Improved YOLOv3 traffic sign detection method based on multi-scale feature layer
Technical Field
The invention relates to the field of computer vision, in particular to an improved YOLOv3 traffic sign detection method based on a multi-scale feature layer.
Background
With the development of science and technology and the improvement of living standard of people, more and more automobiles are produced, the automobiles seem to be a travel tool for travel of people, the life of people is more convenient, and traffic jam and traffic accidents are brought. In order to reduce the occurrence of the matters, an intelligent traffic system is presented, and the traffic sign identification is an important component of the intelligent traffic system, plays an important role in the road driving process of people, and has the working principle that an in-vehicle camera is used for collecting traffic sign pictures from an outdoor complex scene, then a corresponding area is extracted, the traffic sign is detected from the background, and then the characteristics are extracted and classified and identified.
For the research on the identification of the traffic sign, the traditional image identification method is mainly adopted in the early stage, although the traditional image identification method has some effects, the traditional image identification method also has many problems, and the false detection rate and the missed detection rate indexes are relatively high. In recent years, convolutional neural networks are increasingly applied to the field of target detection and target classification, and good research results are obtained, so that people have further breakthrough in the aspect of traffic sign identification.
The patent (CN201910474058.2) applies F-RCNN to detect traffic signs. The patent (CN201910443948.7) applies a convolutional neural network to detect traffic signs. A combination of attention mechanism and neural network for detecting traffic signs is proposed in the patent (CN 201910365006.1). The patent (CN201910097579.0) applies a multi-window traffic sign detection. The YOLOv3 network model is not used in the above patents to detect traffic signs.
The backbone network Darknet53 of YOLOv3 is added with a plurality of layers of volume blocks and residual error networks to form a deeper network structure on the basis of the previous generation Darknet 19. The detection accuracy is improved at the expense of a certain detection speed, but it is still difficult to detect small traffic sign objects.
In the patent (CN201911422311.6), an improved traffic sign detection and identification method based on YOLOv3 is disclosed, which includes the following steps: (1) acquiring and labeling a traffic sign image data set as a training set; (2) constructing a YOLOv3 improved network model; (3) training the Yolov3 improved network model through a training set; (4) and (3) inspecting and identifying the traffic sign image to be detected through the trained YOLOv3 improved network model. The method mainly aims to solve the problems that the size of a YOLOv3 model is optimized, the traffic sign detection effect is poor when the traffic sign is small, and the sample category is unbalanced.
A traffic sign detection and identification method based on YOLOv3 is disclosed in a patent (CN201910131881.3), and the method firstly makes a traffic sign data set according to a VOC format; then, improving the VGG16 network, modifying the number of nodes of a first layer of full connection layer, deleting a second layer of full connection layer, adding a residual error layer in the network, changing an activation function ReLU into a PReLU, changing all Max points in the network into stochastic points, and replacing a darknet53 network in YOLOv3 with the improved network; finally, the data set is trained by using the improved YOLOv3, and the trained model is used for detecting the traffic sign. By adopting the method, the detection capability of YOLOv3 on target scale inconsistency cannot be enhanced.
Disclosure of Invention
The invention aims to provide an improved traffic sign detection method of YOLOv3 based on a multi-scale feature layer, which has the advantages of balanced sample types, strong detection capability for target scale inconsistency and capability of detecting smaller traffic signs.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a traffic sign detection method based on improved YOLOv3 of a multi-scale feature layer comprises the following steps:
step 1, preparing a traffic sign data set and performing data enhancement;
step 2, building a YOLOv3 improved network model, improving a backbone network, replacing the original Darknet53 with a Densenet network, and optimizing the Densenet network;
step 3, training the improved network model of YOLOv 3;
and 4, detecting the traffic sign by using the training optimal model.
The technical scheme of the invention is further improved as follows: in step 1, 45-type traffic signs with the use frequency exceeding 100 times are adopted as a data set.
The technical scheme of the invention is further improved as follows: the step 1 is divided into the following 2 steps:
(1) adopting TT100K to disclose a data set, dividing a training set and a testing set according to the proportion of 8: 2 to the data set, and converting the label format of the data set into VOC by json;
(2) in the generating data enhancement method, firstly, random data enhancement is carried out on a standard template of a traffic sign with a small number of classes in a training set according to a set probability, then the enhanced standard template is added into a background image without the traffic sign to obtain a synthetic image of the class, class marking is carried out, and the synthetic image and an original training set are used as training data.
The technical scheme of the invention is further improved as follows: improvements to the backbone network include: and further extracting features from the four-layer feature map through average pooling of the backbone network each time, then performing up-sampling to fuse the up-sampling with the original features, and outputting feature maps of four scales.
The technical scheme of the invention is further improved as follows: the improvement made to the backbone network further comprises: and adding a spatial pooling pyramid module, pooling the output feature maps by adopting three different pooling checks, and merging the three pooled feature maps and the original input channel.
The technical scheme of the invention is further improved as follows: the improvement made to the backbone network further comprises: and optimizing a Loss function by adopting a GIoU algorithm and a Focal local algorithm, wherein the calculation formula of the GIoU is as follows:
Figure BDA0002851219360000031
LGIoU＝1-GIoU
the meaning of this formula is: finding a minimum occlusion region C can include A and BCalculating the ratio of the area of C without A and B to the total area of C, and finally subtracting the ratio from IoU, LGIoUIs a bounding box loss function;
the calculation formula of Focal local is as follows:
FL(pt)＝-at(1-pt)ylog(pt)
in the formula: y is 2, atThe value is 0.25, and p is the probability that the model predicts the sample to be positive.
The technical scheme of the invention is further improved as follows: and clustering the sizes of the traffic sign data sets by adopting a k-means + + algorithm to generate 12 different sizes as the anchor sizes of the training set.
The technical scheme of the invention is further improved as follows: the optimization of the densenert network specifically comprises the following steps: convolution pooling parallel replaced the densinet initialized 7 × 7 volume Block with the Stem Block module.
The technical scheme of the invention is further improved as follows: the training process in the step 3 specifically comprises the following steps: initializing parameters in the network, inputting the processed training data set into the network in batches for forward propagation and continuously calculating loss, performing backward propagation through a loss function to update the parameters in the network, and storing the network parameters at the moment as a model, wherein the loss value tends to be stable after multiple iterations.
Due to the adoption of the technical scheme, the invention has the technical progress that:
the combination of Desnenet and Stem Block is adopted to replace Darknet53, and the connection between feature layers is enriched.
The four output feature layers are adopted to detect the targets with different sizes, the four scale feature layers are output to extract the targets with different sizes, the spatial pooling pyramid module is adopted, three pooling layers with different sizes are connected in parallel, and the feature layers with different scales are fused, so that the detection capability of the small-size targets is enhanced, and the detection capability of the YOLOv3 on different target scales is also enhanced.
A Loss function is optimized by a GIoU algorithm and a Focal local algorithm, the detection precision of the model is further improved, the size of the anchor is re-clustered by k-means + + aiming at a traffic sign data set, and the problem of unbalanced sample categories is relieved by data enhancement.
Drawings
FIG. 1 is a general framework diagram of the improved YOLOv3 network of the present invention;
FIG. 2 is a Densenet original network;
FIG. 3 is a Dense Block module;
FIG. 4 is a Stem Block module;
FIG. 5 is a block diagram of a spatial pooling pyramid module;
FIG. 6 is a graph of generating data enhancement pre-frequent for a data set;
FIG. 7 is a graph of frequency after generative data enhancement on a data set.
Detailed Description
The present invention will be described in further detail with reference to the following examples:
example one
A traffic sign detection method based on improved YOLOv3 of a multi-scale feature layer comprises the following steps:
step 1, preparing a traffic sign data set and performing data enhancement, wherein the method specifically comprises the following two steps:
(1) the TT100K is adopted to disclose a data set, wherein the training set comprises 6103 pictures in total, the testing set comprises 3067 pictures in total, the picture resolution is 2048 × 2048 high-definition pictures, and at this time, if some traffic sign categories appear too infrequently, the network can hardly learn the characteristics of the traffic sign categories. Therefore, the present patent uses 45-class traffic signs with frequency over 100 times for training. And, to learn enough features, a data set is taken to be 8: 2 into training and test sets. In this case, the tag format of the data set is json file, and manual conversion into the VOC format is required to enable the data set to be used in a network.
(2) In the generating data enhancement method, firstly, random data enhancement is carried out on a standard template of a traffic sign with a small number of classes in a training set according to a set probability, then the enhanced standard template is added into a background image without the traffic sign to obtain a synthetic image of the class, corresponding software is used for carrying out class marking, and the synthetic image and the original training set are used as training data together. The class frequency distribution histograms before and after processing are shown in fig. 6 and 7 by data enhancement to alleviate the problem of sample class imbalance.
And 2, building a YOLOv3 improved network model.
In the present invention, fig. 1 is a general framework diagram of the YOLOv3 improved network. Firstly, the YOLOv3 backbone network is improved, Darknet is replaced by an improved Densenet network, namely a Stem Block convolutional pooling module is used for parallel replacement of a Densenet initialization 7 x 7 volume Block. Desneet is combined with Stem Block to replace Darknet53 to enrich the connection between feature layers, and the structure of the Desneet improvement module is shown in figure 2, figure 3 and figure 4.
And removing the full connection of the last layer of the Densenet, outputting to a subsequent characteristic layer, performing 2 times of upsampling by using a 4 times of downsampled characteristic diagram and an 8 times of downsampled characteristic diagram, then performing characteristic fusion, outputting a fourth characteristic layer for detecting an ultra-small object, and modifying the original multi-scale input into a fixed 608 input size for detecting a small target traffic sign.
After 32 times of down sampling is performed on the backbone network, a spatial pooling pyramid module is added, the structure of which is shown in fig. 5, the spatial pooling pyramid module uses three maximum pooling check feature maps with different sizes to perform pooling respectively, the sizes of the maximum pooling checks are 5 × 5, 9 × 9 and 13 × 13 respectively, and the input filled size padding is as follows:
padding＝(kernelsize-1)/2
and then, channel merging is carried out on the result after the pooling and the original input, so that the detection capability of YOLOv3 on different target scales is enhanced.
And a GIoU algorithm is used as a boundary box Loss function, and the Focal local solves the problem of unbalance of positive and negative samples in a prediction box, so that the detection precision of the model is further improved.
Wherein, the calculation formula of the GIoU is as follows:
LGIoU＝1-GIoU
the meaning of this formula is: finding a minimum occlusion region C can include A and B, calculating the ratio of the area of C without A and B to the total area of C, and subtracting this ratio, L, from IoUGIoUIs a bounding box penalty function.
The calculation formula of Focal local is as follows:
FL(pt)＝-at(1-pt)ylog(pt)
in the formula: y is 2, atThe value is 0.25, and p is the probability that the model predicts the sample to be positive.
A K-means + + algorithm is used for clustering instead of a K-means algorithm used in a YOLOv3 original text, the K-means algorithm randomly selects K points as clustering centers once, the result is influenced by the selection of initial points, the K-means + + algorithm randomly selects a first clustering center, then a point far away from the clustering center is selected as a new clustering center, and the like, 12 anchor values serving as models are selected, and through the method, the K-means + + can effectively improve the prediction error.
And 3, training the improved network model of YOLOv 3.
The picture input size is set to 608, the initial learning rate is set to 1e-3, the processed training data set is input into the network in batches to carry out forward propagation and continuously calculate loss, backward propagation is carried out through a loss function to update various parameters in the network, the loss value tends to be stable after multiple iterations, and the network parameters at the moment are stored as a model.
And 4, detecting the traffic sign by using the training optimal model.
The traffic sign test set is tested by using the optimal training model, so that the precision is greatly improved, and the detection effect is more excellent particularly for detecting small objects.
Example two
A traffic sign detection method based on improved YOLOv3 of a multi-scale feature layer comprises the following steps:
step 1, preparing a traffic sign data set and performing data enhancement, wherein the method specifically comprises the following two steps:
(1) the TT100K was used to disclose a data set, where the training set contained 6103 pictures, the test set 3067 pictures, and the resolution of the pictures was 2048 × 2048 high definition pictures, at which time the training set and the test set were partitioned at an 8: 2 ratio to learn enough features. In this case, the tag format of the data set is json file, and manual conversion into the VOC format is required to enable the data set to be used in a network.
(2) In the generating data enhancement method, firstly, random data enhancement is carried out on a standard template of a traffic sign with a small number of classes in a training set according to a set probability, then the enhanced standard template is added into a background image without the traffic sign to obtain a synthetic image of the class, corresponding software is used for carrying out class marking, and the synthetic image and the original training set are used as training data together. The class frequency distribution histograms before and after processing are shown in fig. 6 and 7 by data enhancement to alleviate the problem of sample class imbalance.
And 2, building a YOLOv3 improved network model.
In the present invention, fig. 1 is a general framework diagram of the YOLOv3 improved network. Firstly, the YOLOv3 backbone network is improved, Darknet is replaced by an improved Densenet network, namely a Stem Block convolutional pooling module is used for parallel replacement of a Densenet initialization 7 x 7 volume Block. Desneet is combined with Stem Block to replace Darknet53 to enrich the connection between feature layers, and the structure of the Desneet improvement module is shown in figure 2, figure 3 and figure 4.
And removing the full connection of the last layer of the Densenet, outputting to a subsequent characteristic layer, performing 2 times of upsampling by using a 4 times of downsampled characteristic diagram and an 8 times of downsampled characteristic diagram, then performing characteristic fusion, outputting a fourth characteristic layer for detecting an ultra-small object, and modifying the original multi-scale input into a fixed 608 input size for detecting a small target traffic sign.
After 32 times of down sampling is performed on the backbone network, a spatial pooling pyramid module is added, the structure of which is shown in fig. 5, the spatial pooling pyramid module uses three maximum pooling check feature maps with different sizes to perform pooling respectively, the sizes of the maximum pooling checks are 5 × 5, 9 × 9 and 13 × 13 respectively, and the input filled size padding is as follows:
padding＝(kernelsize-1)/2
and then, channel merging is carried out on the result after the pooling and the original input, so that the detection capability of YOLOv3 on different target scales is enhanced.
And a GIoU algorithm is used as a boundary box Loss function, and the Focal local solves the problem of unbalance of positive and negative samples in a prediction box, so that the detection precision of the model is further improved.
Wherein, the calculation formula of the GIoU is as follows:
Figure BDA0002851219360000081
LGIou＝1-GIoU
the meaning of this formula is: finding a minimum occlusion region C can include A and B, calculating the ratio of the area of C without A and B to the total area of C, and subtracting this ratio, L, from IoUGIoUIs a bounding box penalty function.
The calculation formula of Focal local is as follows:
Figure BDA0002851219360000091
FL(pt)＝-at(1-pt)ylog(pt)
in the formula: y is 2, atThe value is 0.25, and p is the probability that the model predicts the sample to be positive.
A K-means + + algorithm is used for clustering instead of a K-means algorithm used in a YOLOv3 original text, the K-means algorithm randomly selects K points as clustering centers once, the result is influenced by the selection of initial points, the K-means + + algorithm randomly selects a first clustering center, then a point far away from the clustering center is selected as a new clustering center, and the like, 12 anchor values serving as models are selected, and through the method, the K-means + + can effectively improve the prediction error.
And 3, training the improved network model of YOLOv 3.
The picture input size is set to 608, the initial learning rate is set to 1e-3, the processed training data set is input into the network in batches to carry out forward propagation and continuously calculate loss, backward propagation is carried out through a loss function to update various parameters in the network, the loss value tends to be stable after multiple iterations, and the network parameters at the moment are stored as a model.
And 4, detecting the traffic sign by using the training optimal model.
The traffic sign test set is tested by using the optimal training model, so that the precision is greatly improved, and the detection effect is more excellent particularly for detecting small objects.
The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solution of the present invention by those skilled in the art should fall within the protection scope defined by the claims of the present invention without departing from the spirit of the present invention.

Claims (9)

1. A traffic sign detection method based on improved YOLOv3 of a multi-scale feature layer is characterized by comprising the following steps:
step 1, preparing a traffic sign data set and performing data enhancement;
step 2, building a YOLOv3 improved network model, improving a backbone network, replacing the original Darknet53 with a Densenet network, and optimizing the Densenet network;
step 3, training the improved network model of YOLOv 3;
and 4, detecting the traffic sign by using the training optimal model.
2. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 1, wherein: in step 1, 45-type traffic signs with the use frequency exceeding 100 times are adopted as a data set.
3. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 1, wherein: the step 1 is divided into the following 2 steps:
(1) the TT100K was used to disclose the data set, and 8: 2, dividing a training set and a test set in proportion, and converting the label format of the data set into VOC (volatile organic compounds) by json;
(2) in the generating data enhancement method, firstly, random data enhancement is carried out on a standard template of a traffic sign with a small number of classes in a training set according to a set probability, then the enhanced standard template is added into a background image without the traffic sign to obtain a synthetic image of the class, class marking is carried out, and the synthetic image and an original training set are used as training data.
4. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 3, wherein: improvements to the backbone network include: and further extracting features from the four-layer feature map through average pooling of the backbone network each time, then performing up-sampling to fuse the up-sampling with the original features, and outputting feature maps of four scales.
5. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 4, wherein: the improvement made to the backbone network further comprises: and adding a spatial pooling pyramid module, pooling the output feature maps by adopting three different pooling checks, and merging the three pooled feature maps and the original input channel.
6. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 5, wherein: the improvement made to the backbone network further comprises: and optimizing a Loss function by adopting a GIoU algorithm and a Focal local algorithm, wherein the calculation formula of the GIoU is as follows:
LGIoU＝1-GIoU
the meaning of this formula is: finding a minimum occlusion region C can include A and B, calculating the ratio of the area of C without A and B to the total area of C, and finally subtracting this ratio, L, from IoUGIoUIs a bounding box loss function;
the calculation formula of Focal local is as follows:
Figure FDA0002851219350000022
FL(pt)＝-at(1-pt)ylog(pt)
in the formula: y is 2, atThe value is 0.25, and p is the probability that the model predicts the sample to be positive.
7. The method as claimed in claim 6, wherein the traffic sign detection method based on improved YOLOv3 with multi-scale feature layers is characterized in that: the improvement made to the backbone network further comprises: and clustering the sizes of the traffic sign data sets by adopting a k-means + + algorithm to generate 12 different sizes as the anchor sizes of the training set.
8. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 7, wherein: the optimization of the densenert network specifically comprises the following steps: convolution pooling parallel replaced the densinet initialized 7 × 7 volume Block with the Stem Block module.
9. The method for detecting traffic signs based on improved YOLOv3 with multi-scale feature layers as claimed in claim 8, wherein: the training process in the step 3 specifically comprises the following steps: initializing parameters in the network, inputting the processed training data set into the network in batches for forward propagation and continuously calculating loss, performing backward propagation through a loss function to update the parameters in the network, and storing the network parameters at the moment as a model, wherein the loss value tends to be stable after multiple iterations.
CN202011527413.7A 2020-12-22 2020-12-22 Improved YOLOv3 traffic sign detection method based on multi-scale feature layer Pending CN112528934A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011527413.7A CN112528934A (en) 2020-12-22 2020-12-22 Improved YOLOv3 traffic sign detection method based on multi-scale feature layer

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011527413.7A CN112528934A (en) 2020-12-22 2020-12-22 Improved YOLOv3 traffic sign detection method based on multi-scale feature layer

Publications (1)

Publication Number Publication Date
CN112528934A true CN112528934A (en) 2021-03-19

Family

ID=75002382

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011527413.7A Pending CN112528934A (en) 2020-12-22 2020-12-22 Improved YOLOv3 traffic sign detection method based on multi-scale feature layer

Country Status (1)

Country Link
CN (1) CN112528934A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435269A (en) * 2021-06-10 2021-09-24 华东师范大学 Improved water surface floating object detection and identification method and system based on YOLOv3
CN114092917A (en) * 2022-01-10 2022-02-25 南京信息工程大学 MR-SSD-based shielded traffic sign detection method and system
CN114463772A (en) * 2022-01-13 2022-05-10 苏州大学 Deep learning-based traffic sign detection and identification method and system
CN114881992A (en) * 2022-05-24 2022-08-09 北京安德医智科技有限公司 Skull fracture detection method and device and storage medium
CN115147348A (en) * 2022-05-05 2022-10-04 合肥工业大学 Improved YOLOv 3-based tire defect detection method and system

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109685152A (en) * 2018-12-29 2019-04-26 北京化工大学 A kind of image object detection method based on DC-SPP-YOLO
CN110210362A (en) * 2019-05-27 2019-09-06 中国科学技术大学 A kind of method for traffic sign detection based on convolutional neural networks
CN110766098A (en) * 2019-11-07 2020-02-07 中国石油大学(华东) Traffic scene small target detection method based on improved YOLOv3
CN110795991A (en) * 2019-09-11 2020-02-14 西安科技大学 Mining locomotive pedestrian detection method based on multi-information fusion
CN111274970A (en) * 2020-01-21 2020-06-12 南京航空航天大学 Traffic sign detection method based on improved YOLO v3 algorithm

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109685152A (en) * 2018-12-29 2019-04-26 北京化工大学 A kind of image object detection method based on DC-SPP-YOLO
CN110210362A (en) * 2019-05-27 2019-09-06 中国科学技术大学 A kind of method for traffic sign detection based on convolutional neural networks
CN110795991A (en) * 2019-09-11 2020-02-14 西安科技大学 Mining locomotive pedestrian detection method based on multi-information fusion
CN110766098A (en) * 2019-11-07 2020-02-07 中国石油大学(华东) Traffic scene small target detection method based on improved YOLOv3
CN111274970A (en) * 2020-01-21 2020-06-12 南京航空航天大学 Traffic sign detection method based on improved YOLO v3 algorithm

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ROBERT J. WANG等: ""Pelee: A Real-Time Object Detection System on Mobile Devices"", 《ARXIV》 *
张广世等: ""基于改进 YOLOv3网络的齿轮缺陷检测"", 《激光与光电子学进展》 *
李成跃等: ""基于改进YOLO 轻量化网络的目标检测方法"", 《激光与光电子学进展》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435269A (en) * 2021-06-10 2021-09-24 华东师范大学 Improved water surface floating object detection and identification method and system based on YOLOv3
CN114092917A (en) * 2022-01-10 2022-02-25 南京信息工程大学 MR-SSD-based shielded traffic sign detection method and system
CN114463772A (en) * 2022-01-13 2022-05-10 苏州大学 Deep learning-based traffic sign detection and identification method and system
CN115147348A (en) * 2022-05-05 2022-10-04 合肥工业大学 Improved YOLOv 3-based tire defect detection method and system
CN114881992A (en) * 2022-05-24 2022-08-09 北京安德医智科技有限公司 Skull fracture detection method and device and storage medium

Similar Documents

Publication Publication Date Title
CN112528934A (en) Improved YOLOv3 traffic sign detection method based on multi-scale feature layer
CN108830285B (en) Target detection method for reinforcement learning based on fast-RCNN
CN110929577A (en) Improved target identification method based on YOLOv3 lightweight framework
CN106557778A (en) Generic object detection method and device, data processing equipment and terminal device
CN105574550A (en) Vehicle identification method and device
CN107945153A (en) A kind of road surface crack detection method based on deep learning
WO2022083784A1 (en) Road detection method based on internet of vehicles
CN111967313B (en) Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm
CN112016605B (en) Target detection method based on corner alignment and boundary matching of bounding box
CN113256649B (en) Remote sensing image station selection and line selection semantic segmentation method based on deep learning
CN111191608B (en) Improved traffic sign detection and identification method based on YOLOv3
CN109886147A (en) A kind of more attribute detection methods of vehicle based on the study of single network multiple-task
CN110210431B (en) Point cloud semantic labeling and optimization-based point cloud classification method
CN110222604A (en) Target identification method and device based on shared convolutional neural networks
CN111832615A (en) Sample expansion method and system based on foreground and background feature fusion
CN109189965A (en) Pictograph search method and system
CN113076804B (en) Target detection method, device and system based on YOLOv4 improved algorithm
CN110728307A (en) Method for realizing small sample character recognition of X-ray image by self-generating data set and label
CN111898570A (en) Method for recognizing text in image based on bidirectional feature pyramid network
CN113159024A (en) License plate recognition technology based on improved YOLOv4
CN111612803A (en) Vehicle image semantic segmentation method based on image definition
CN107169450A (en) The scene classification method and system of a kind of high-resolution remote sensing image
CN113963333B (en) Traffic sign board detection method based on improved YOLOF model
CN113327227B (en) MobileneetV 3-based wheat head rapid detection method
CN114241470A (en) Natural scene character detection method based on attention mechanism

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