CN115311647B - Car logo detection and identification method fusing car logo classification features - Google Patents
Car logo detection and identification method fusing car logo classification features Download PDFInfo
- Publication number
- CN115311647B CN115311647B CN202211223901.8A CN202211223901A CN115311647B CN 115311647 B CN115311647 B CN 115311647B CN 202211223901 A CN202211223901 A CN 202211223901A CN 115311647 B CN115311647 B CN 115311647B
- Authority
- CN
- China
- Prior art keywords
- logo
- feature
- car logo
- car
- relu
- 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.)
- Active
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
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Abstract
The invention belongs to the technical field of vehicle logo detection and identification, and particularly relates to a vehicle logo detection and identification method fusing vehicle logo classification features.
Description
Technical Field
The invention belongs to the technical field of vehicle logo detection and identification, and particularly relates to a vehicle logo detection and identification method fusing vehicle logo classification features.
Background
The automatic identification of the car logo is an important component of the automatic identification system of the car attribute, the problem of car logo identification accuracy is solved, the car logo serves as key information of a car and is not easy to replace, the car logo can become a significant feature of the car, if the car logo can be accurately positioned, the accuracy of car classification and identification can be effectively improved, and therefore the detection of the car logo has very important significance for car detection management and control.
The traditional algorithm for positioning and identifying the vehicle logo mainly comprises the steps of firstly finding out the position of the vehicle logo by utilizing the position relation between the vehicle logo and the vehicle license plate, then roughly positioning the vehicle logo above the vehicle license plate, secondly positioning the vehicle logo according to a morphological operator, and then classifying the positioned vehicle logo by utilizing a pattern recognition technology.
The vehicle logo recognition technology based on the detection technology mainly comprises two parts, namely vehicle logo positioning and vehicle logo recognition, wherein a vehicle logo is positioned firstly, then the vehicle logo is recognized, and with the development of a deep learning technology, a vehicle logo detection method based on a deep learning target detection algorithm becomes the mainstream of a vehicle logo detection algorithm. Therefore, an effective means for improving the accuracy and the recognition rate of the car logo detection algorithm is urgently needed.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a vehicle logo detection and identification method integrating vehicle logo classification features, so that the vehicle logo detection precision and identification rate are improved.
In order to achieve the purpose, the invention firstly collects commonly used car logo images, designs a car logo classification network, then fuses one-dimensional features obtained by the car logo classification network into a decoding layer of the car logo segmentation network of the car image after copying and expanding, and enables the car logo features of the car image to generate maximum correspondence when being similar to the car logo features in the classification network, so that a car area can be highlighted in the image, the segmentation of the car logo area is completed, and the maximum circumscribed rectangle of the car logo area is the detected and identified car logo, and the invention specifically comprises the following steps:
(1) Collecting 80 types of car logo images to form a car logo data set, wherein 500 images are collected for each type of car logo, 40000 images are collected in total, and the car logo images only contain the car logos and do not contain other contents;
(2) The method comprises the steps of scaling a car logo image to 64 x 3 pixel size, inputting the car logo image into a car logo classification network, processing the car logo image through four convolution layers, RELU (batch standardization), BN (batch standardization) and a pooling layer to obtain a 4 x 512 feature image, processing the car logo image through two full connection layers to obtain 80-dimensional car logo classification features, and enabling a car logo data set to be 8:1:1, dividing the ratio of the first full connection layer into a training set, a verification set and a test set, training 300 iteration termination storage parameter models, and taking 512 features B1 output by the first full connection layer as classification features of the vehicle for assisting in segmenting the vehicle logo image;
(3) Scaling the car logo image to 512 × 512 × 3 pixels, inputting the scaled car logo image into a car logo segmentation network, and processing the car logo image and 512 features B1 output by the first full connection layer in the step (2) together to obtain a segmentation result;
(4) Training the car logo segmentation network: 20000 images containing different car logos are collected and marked by using an image segmentation marking tool, and the data set is as follows: 1:1, dividing the energy function of the segmentation network into a training set, a verification set and a test set, wherein the energy function of the segmentation network is a cross entropy loss function, fixing the characteristic after B1 down-sampling during training, namely the classification characteristic of the car logo is not changed along with the training of the segmentation network, training 800 iteration termination storage parameter models, and obtaining a trained network;
(5) Testing the car logo segmentation network: and (4) scaling the image to be tested to 512 multiplied by 3, sending the image to the network trained in the step (3) to obtain the segmentation result of the car logo, wherein the maximum circumscribed rectangle of the segmentation result is the detected and identified car logo.
As a further technical solution of the present invention, the four convolution, RELU, BN (batch normalization) and pooling layers in step (2) are a convolution layer, BN (batch normalization), RELU and average pooling layer containing 64 convolutions of 3 × 3 × 3, respectively; convolutional layer, BN (batch normalization), RELU, and average pooling layer containing 128 3 × 3 convolutions; convolutional layer containing 256 3 × 3 convolutions, BN (batch normalization), RELU, and average pooling layer; convolutional layer containing 512 3 × 3 convolutions, BN (batch normalization), RELU, and average pooling layer.
In a further technical solution of the present invention, in the step (2), the number of the nerve units in the first full junction layer is 512, and the number of the nerve units in the second full junction layer is 80.
As a further technical scheme of the invention, the energy loss of the car logo classification network in the step (2) adopts a cross entropy loss function.
As a further technical scheme of the invention, the specific process of the step (3) is as follows:
(31) Sequentially performing convolution layer operation, BN (batch standardization), RELU (remote unified planning) and pooling layer operation on an automobile logo image input into an automobile logo segmentation network, outputting a feature S1 with a feature dimension of 256 × 256 × 64, expanding the B1 feature dimension obtained in the step (2) to 256 × 256 × 64 (reducing the feature dimension to 64 dimensions by utilizing down sampling, expanding the feature in a plane to enable the feature of each position to be a B1 dimension reduction feature), representing the feature by a symbol B1E, and performing series operation on the S1 and the B1E to obtain a feature with a feature dimension of 256 × 256 × 128;
(32) Inputting the features obtained in the step (31) into a convolutional layer containing 128 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operation, outputting a feature S2 with a feature dimension of 128 × 128 × 128 (feature dimension is reduced to 128 dimensions by down-sampling, and the feature dimension of each position is expanded in a plane so that the feature of each position is a dimension-reduced B1 feature), and obtaining a feature with a feature dimension of 128 × 128 × 256 by performing tandem operation on S2 and B2E, as indicated by symbol B2E;
(33) Inputting the features obtained in the step (32) into a convolutional layer containing 256 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operation, outputting a feature S3 with a feature dimension of 64 × 64 × 256, expanding the B1 feature dimension obtained in the step (2) to 64 × 64 × 256 (reducing the feature dimension to 256 dimensions by down-sampling, and expanding the feature in a plane so that the feature at each position is a feature with a dimension reduced by B1), and obtaining a feature with a feature dimension of 64 × 64 × 512 by performing a concatenation operation of S3 and B3E, as indicated by symbol B3E;
(34) Inputting the features obtained in the step (33) into an deconvolution layer containing 256 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operations, and outputting a feature S4 having a feature dimension of 128 × 128 × 256;
(35) Inputting the features obtained in the step (34) into an deconvolution layer containing 128 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operations, and outputting a feature S5 having a feature dimension of 256 × 256 × 128;
(36) Inputting the features obtained in step (35) into an deconvolution layer containing 64 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operations, and outputting a feature dimension of 512 × 512 × 64, and obtaining a 512 × 512 × 1 segmentation result after 1 × 1 convolution.
Compared with the prior art, the method collects the commonly used car logo images, designs the car logo classification network, fuses the one-dimensional features obtained by the car logo classification network into the decoding layer of the car logo segmentation network of the car image after copying and expanding, and enables the car logo features of the car image to generate maximum correspondence when being similar to the car logo features in the classification network, so that the car area can be highlighted in the image, the segmentation of the car logo area is completed, the maximum circumscribed rectangle of the car logo area is the detected and identified car logo, the method can well solve the car logo image identification problem, and the accuracy and the identification rate of car logo detection are improved.
Drawings
Fig. 1 is a diagram of a car logo image segmentation network structure integrating car logo classification features according to the invention.
Detailed Description
The invention will be further described by way of examples, without in any way limiting the scope of the invention, with reference to the accompanying drawings.
Example (b):
the embodiment provides a car logo detection and recognition method fusing car logo classification features, wherein the car logo features are obtained through a training car logo classification network and then fused to a car logo segmentation network to obtain car logo segmentation results, so that the detection and classification of car logos are realized. Fig. 1 is a diagram of a car logo image segmentation network structure integrating car logo classification features, and as shown in fig. 1, the specific implementation includes the following steps:
(1) Collecting a car logo data set and designing a car logo classification network:
and collecting a data set formed by 80 types of car logos, collecting 500 images of each type of car logo, wherein 40000 images are total, and the car logo image only contains the car logo and does not contain other contents, namely the car logo is extracted from the car image when a sample is collected.
Designing a car logo classification network, wherein the structure diagram of the car logo classification network is shown in the lower half of the attached drawing 1 of the specification, a car logo image is scaled to 64 multiplied by 3 pixels and is input into the car logo classification network as an input image, the input image is subjected to four convolution, RELU and pooling layer operations (respectively: 64 convolution layers containing 3 multiplied by 3 convolution, BN (batch standardization), RELU and average pooling layer; 128 convolution layers containing 3 multiplied by 3 convolution, BN (batch standardization), RELU and average pooling layer; 256 convolution layers containing 3 multiplied by 3 convolution, BN (batch standardization), RELU and average pooling layer; 512 convolution layers containing 3 multiplied by 3 convolution, BN (batch standardization), RELU and average pooling layer) to obtain a 4 multiplied by 512 characteristic image, and 80 dimensional car logo classification characteristics are obtained after passing through two full connection layers (512 neural units of a first full connection layer and 80 neural units of a second full connection layer), and the energy loss of the network adopts an entropy loss function;
the car logo data set is as follows 8:1:1, dividing the ratio of the first full connection layer into a training set, a verification set and a test set, training 300 iteration termination storage parameter models, and taking 512 features B1 output by the first full connection layer as classification features of the vehicle for assisting in segmenting the vehicle logo image;
(2) Designing a car logo segmentation network fusing car logo classification features:
designing a car logo segmentation network fusing car logo classification features as shown in the upper half of the attached drawing 1 of the specification, an image containing a car logo is input into the network as an input image, the input image is subjected to convolutional layer containing 64 3 × 3 × 3 convolutions, BN (batch normalization), RELU and pooling layer operation, the output feature dimension is 256 × 256 × 64 and is represented by a symbol S1, the B1 feature dimension obtained in step 1) is expanded to 256 × 256 × 64 (the feature dimension is reduced to 64 dimensions by down-sampling and is expanded in a plane so that the feature of each position is a B1 dimension-reduced feature), the feature dimension after the series operation of S1 and B1E is 256 × 256 × 128 is represented by a symbol B1E, the feature is input into the convolutional, BN (batch normalization), RELU and pooling layer operation containing 128 convolutions 3 × 3 convolutions, the output feature dimension is 128 × 128 × 128, represented by symbol S2, the B1 feature dimension obtained in step 1) is expanded to 128 × 128 × 128 (the feature dimension is reduced to 128 dimensions by down-sampling, and expanded in a plane so that the feature of each position is a feature of B1 dimension reduction), represented by symbol B2E, the feature dimension after the S2 and B2E are connected in series is 128 × 128 × 256, the feature is input to a convolutional layer containing 256 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operation, the output feature dimension is 64 × 64 × 256, represented by symbol S3, the B1 feature dimension obtained in step 1) is expanded to 64 × 64 × 256 (the feature dimension is reduced to 256 dimensions by down-sampling, and expanded in a plane so that the feature of each position is a feature of B1 dimension reduction), denoted by symbol B3E, the feature dimension after the series operation of S3 and B3E is 64 × 64 × 512, the feature is input to the deconvolution layer containing 256 convolutions of 3 × 3, BN (batch normalization), RELU, pooling layer operation, the output feature dimension is 128 × 128 × 256, denoted by symbol S4, the feature is input to the deconvolution layer containing 128 convolutions of 3 × 3, BN (batch normalization), RELU, pooling layer operation, the output characteristic dimension is 256 × 256 × 128, indicated by symbol S5, and is input to the deconvolution layer containing 64 3 × 3 convolutions, BN (batch normalization), RELU, pooling layer operations, the output characteristic dimension is 512 × 512 × 64, and a segmentation result of 512 × 512 × 1 is obtained after 1 × 1 convolution;
(3) Training a car logo segmentation network:
20000 images containing different car logos are collected and marked by using an image segmentation and marking tool, and the data set is as follows: 1:1, dividing the energy function of the segmentation network into a training set, a verification set and a test set, fixing the characteristic after B1 down-sampling during training, namely the classification characteristic of the car logo is not changed along with the training of the segmentation network, training 800 iteration termination storage parameter models, and obtaining the trained car logo segmentation network;
(4) Testing the car logo segmentation network:
inputting a test image, scaling the test image to 512 multiplied by 3, sending the test image to the network trained in the step (3), and obtaining the segmentation result of the car logo, wherein the maximum circumscribed rectangle of the segmentation result is the detected and identified car logo area.
Network architectures and algorithms not described in detail herein are all common in the art.
It is noted that the disclosed embodiments are intended to aid in further understanding of the invention, but those skilled in the art will appreciate that: various substitutions and modifications are possible without departing from the spirit and scope of this disclosure and the appended claims. Therefore, the invention should not be limited by the disclosure of the embodiments, but should be defined by the scope of the appended claims.
Claims (5)
1. A car logo detection and identification method fused with car logo classification features is characterized by comprising the following steps:
(1) Collecting 80 types of car logo images to form a car logo data set, wherein 500 images are collected for each type of car logo, 40000 images are collected, and the car logo images only contain car logos and do not contain other contents;
(2) The vehicle logo image is input into a vehicle logo classification network after being scaled to 64 multiplied by 3 pixel size, the vehicle logo image is processed by four convolution, RELU, batch standardization and pooling layers to obtain 4 multiplied by 512 feature image, and is processed by two full connection layers to obtain 80-dimensional vehicle logo classification features, and the vehicle logo data set is processed according to the following steps of 8:1:1, dividing the ratio of the first full connection layer into a training set, a verification set and a test set, training 300 iteration termination storage parameter models, and taking 512 features B1 output by the first full connection layer as classification features of the vehicle for assisting in segmenting the vehicle logo image;
(3) Scaling the car logo image to 512 × 512 × 3 pixels, inputting the scaled car logo image into a car logo segmentation network, and processing the car logo image and 512 features B1 output by the first full connection layer in the step (2) together to obtain a segmentation result;
(4) Training the car logo segmentation network: 20000 images containing different car logos are collected and marked by using an image segmentation and marking tool, and the data set is as follows: 1:1, dividing the energy function of the segmentation network into a training set, a verification set and a test set, wherein the energy function of the segmentation network is a cross entropy loss function, fixing the characteristic after B1 down-sampling during training, namely the classification characteristic of the car logo is not changed along with the training of the segmentation network, training 800 iteration termination storage parameter models, and obtaining a trained network;
(5) Testing the car logo segmentation network: and (4) scaling the image to be tested to 512 multiplied by 3, sending the image to the network trained in the step (4) to obtain the segmentation result of the car logo, wherein the maximum circumscribed rectangle of the segmentation result is the detected and identified car logo.
2. The method for detecting and identifying the vehicle logo with the fused vehicle logo classification features as claimed in claim 1, wherein the four convolution, RELU, batch normalization and pooling layers in the step (2) are a convolution layer, a batch normalization, a RELU and an average pooling layer containing 64 convolutions of 3 × 3 × 3 respectively; convolutional layers containing 128 3 × 3 convolutions, batch normalization, RELU, and average pooling layers; convolutional layers containing 256 convolutions of 3 × 3, batch normalization, RELU, and average pooling layers; convolutional layers with 512 3 × 3 convolutions, batch normalization, RELU, and average pooling layers.
3. The method for detecting and identifying the emblem integrated with the car logo classification features according to claim 2, wherein in the step (2), the number of the nerve units of the first fully-connected layer of the two fully-connected layers is 512, and the number of the nerve units of the second fully-connected layer is 80.
4. The method for detecting and identifying the car logo fusing the car logo classification features according to claim 3, wherein the energy loss of the car logo classification network in the step (2) adopts a cross entropy loss function.
5. The method for detecting and identifying the car logo fused with the car logo classification features according to claim 4, wherein the specific process of the step (3) is as follows:
(31) Sequentially performing convolution layer containing 64 convolutions of 3 multiplied by 3, batch standardization, RELU and pooling layer operation on the vehicle logo image input into the vehicle logo segmentation network, outputting a feature S1 with a feature dimension of 256 multiplied by 64, expanding the B1 feature dimension obtained in the step (2) to 256 multiplied by 64, representing the B1 feature dimension by a symbol B1E, and performing series operation on the S1 and the B1E to obtain a feature with a feature dimension of 256 multiplied by 128;
(32) Inputting the features obtained in the step (31) into a convolutional layer containing 128 3 × 3 convolutions, performing batch standardization, RELU and pooling layer operation, outputting a feature S2 with a feature dimension of 128 × 128 × 128, expanding the feature dimension of B1 obtained in the step (2) to 128 × 128, using a symbol B2E to represent the feature, and performing tandem operation on the S2 and the B2E to obtain a feature with a feature dimension of 128 × 128 × 256;
(33) Inputting the features obtained in the step (32) into a convolutional layer containing 256 3 × 3 convolutions, performing batch normalization, RELU and pooling layer operation, outputting a feature S3 with a feature dimension of 64 × 64 × 256, expanding the B1 feature dimension obtained in the step (2) to 64 × 64 × 256, representing the feature with a symbol B3E, and performing tandem operation on the S3 and the B3E to obtain a feature with a feature dimension of 64 × 64 × 512;
(34) Inputting the features obtained in the step (33) into an deconvolution layer containing 256 3 × 3 convolutions, performing batch normalization, RELU, and pooling layer operations, and outputting a feature S4 having a feature dimension of 128 × 128 × 256;
(35) Inputting the features obtained in the step (34) into an deconvolution layer containing 128 3 × 3 convolutions, performing batch normalization, RELU and pooling layer operations, and outputting a feature S5 with a feature dimension of 256 × 256 × 128;
(36) Inputting the features obtained in the step (35) into an deconvolution layer containing 64 3 × 3 convolutions, performing batch standardization, RELU and pooling layer operations, outputting a feature dimension of 512 × 512 × 64, and performing 1 × 1 convolution to obtain a 512 × 512 × 1 segmentation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211223901.8A CN115311647B (en) | 2022-10-09 | 2022-10-09 | Car logo detection and identification method fusing car logo classification features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211223901.8A CN115311647B (en) | 2022-10-09 | 2022-10-09 | Car logo detection and identification method fusing car logo classification features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115311647A CN115311647A (en) | 2022-11-08 |
| CN115311647B true CN115311647B (en) | 2023-01-24 |
Family
ID=83867054
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211223901.8A Active CN115311647B (en) | 2022-10-09 | 2022-10-09 | Car logo detection and identification method fusing car logo classification features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115311647B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105868774A (en) * | 2016-03-24 | 2016-08-17 | 西安电子科技大学 | Selective search and convolutional neural network based vehicle logo recognition method |
| CN109359666A (en) * | 2018-09-07 | 2019-02-19 | 佳都新太科技股份有限公司 | A kind of model recognizing method and processing terminal based on multiple features fusion neural network |
| CN109948643A (en) * | 2019-01-21 | 2019-06-28 | 东南大学 | A kind of type of vehicle classification method based on deep layer network integration model |
| CN111104944A (en) * | 2019-12-26 | 2020-05-05 | 中国计量大学 | License plate character detection and segmentation method based on R-FCN |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764306B (en) * | 2018-05-15 | 2022-04-22 | 深圳大学 | Image classification method and device, computer equipment and storage medium |
| CN110837807A (en) * | 2019-11-11 | 2020-02-25 | 内蒙古大学 | Identification method and system for fake-licensed vehicle |
| CN111368909B (en) * | 2020-03-03 | 2021-05-11 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
-
2022
- 2022-10-09 CN CN202211223901.8A patent/CN115311647B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105868774A (en) * | 2016-03-24 | 2016-08-17 | 西安电子科技大学 | Selective search and convolutional neural network based vehicle logo recognition method |
| CN109359666A (en) * | 2018-09-07 | 2019-02-19 | 佳都新太科技股份有限公司 | A kind of model recognizing method and processing terminal based on multiple features fusion neural network |
| CN109948643A (en) * | 2019-01-21 | 2019-06-28 | 东南大学 | A kind of type of vehicle classification method based on deep layer network integration model |
| CN111104944A (en) * | 2019-12-26 | 2020-05-05 | 中国计量大学 | License plate character detection and segmentation method based on R-FCN |
Non-Patent Citations (4)
| Title |
|---|
| "Internet-Vision Based Vehicle Model Query System Using Eigenfaces and Pyramid of Histogram of Oriented Gradients";Thitiphat Anakavej 等;《IEEE》;20140130;179-186 * |
| YOLOv3网络在车标检测中的应用;王林等;《电子技术应用》;20200906(第09期);40-43 * |
| 基于Edge Boxes的大型车辆车标检测与识别;李熙莹等;《计算机工程与应用》;20171010(第12期);157-164 * |
| 基于深度卷积神经网络的车标分类;杨帅等;《工业仪表与自动化装置》;20171015(第05期);77-80 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115311647A (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111292305B (en) | Improved YOLO-V3 metal processing surface defect detection method | |
| CN109190752B (en) | Image semantic segmentation method based on global features and local features of deep learning | |
| Ohya et al. | Recognizing characters in scene images | |
| CN113077444A (en) | CNN-based ultrasonic nondestructive detection image defect classification method | |
| CN113240623A (en) | Pavement disease detection method and device | |
| CN115439458A (en) | Industrial image defect target detection algorithm based on depth map attention | |
| CN114266881A (en) | Pointer type instrument automatic reading method based on improved semantic segmentation network | |
| CN115272204A (en) | Bearing surface scratch detection method based on machine vision | |
| CN110008949B (en) | Image target detection method, system, device and storage medium | |
| CN115311647B (en) | Car logo detection and identification method fusing car logo classification features | |
| CN111539330B (en) | Transformer substation digital display instrument identification method based on double-SVM multi-classifier | |
| CN111612803B (en) | Vehicle image semantic segmentation method based on image definition | |
| CN110287970B (en) | Weak supervision object positioning method based on CAM and covering | |
| Kälber et al. | U-Net based Zero-hour Defect Inspection of Electronic Components and Semiconductors. | |
| CN111950556A (en) | License plate printing quality detection method based on deep learning | |
| CN113392852B (en) | Vehicle detection method and system based on deep learning | |
| CN112257525A (en) | Logistics vehicle card punching identification method, device, equipment and storage medium | |
| CN111553336B (en) | Print Uyghur document image recognition system and method based on link segment | |
| CN117067112B (en) | Water cutting machine and control method thereof | |
| CN114708457A (en) | Hyperspectral deep learning identification method for purple fringing resistance identification | |
| CN115631197A (en) | Image processing method, device, medium, equipment and system | |
| CN115272377A (en) | Vehicle segmentation method fusing image edge information | |
| Ramadhan et al. | Symbol Spotting in Electronic Images Using Morphological Filters and Hough Transform | |
| CN117173105A (en) | Chip defect detection method, device, equipment and medium | |
| CN114187273A (en) | Method and system for detecting multiple types of flaws on inner wall of corrugated pipe based on deep learning |
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 |