CN112085001A - Tunnel recognition model and method based on multi-scale edge feature detection - Google Patents
Tunnel recognition model and method based on multi-scale edge feature detection Download PDFInfo
- Publication number
- CN112085001A CN112085001A CN202011007456.2A CN202011007456A CN112085001A CN 112085001 A CN112085001 A CN 112085001A CN 202011007456 A CN202011007456 A CN 202011007456A CN 112085001 A CN112085001 A CN 112085001A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- training
- tunnel
- feature detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 66
- 238000010276 construction Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 24
- 238000012544 monitoring process Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 description 6
- 238000012795 verification Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
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
-
- 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
Abstract
The invention discloses a tunnel identification model and a method based on multi-scale edge feature detection, wherein the model comprises the following steps: the data set construction module is used for constructing a data set for model training by utilizing images of various driving roads; the training sample loading module is used for loading training samples in batches during training, cutting the image size into the same size during loading, and inputting the image size into the recognition network; the method comprises the steps of identifying a network, constructing the identification network based on a residual error network Resnet34, using an edge binary image as a label when training the network, guiding the network to learn edge characteristics when learning image characteristics, and combining multi-scale edge characteristics to predict a tunnel more accurately.
Description
Technical Field
The invention relates to the technical field of computer image processing, in particular to a tunnel identification model and method based on multi-scale edge feature detection.
Background
At present, the main method for processing road images shot by automobiles in automatic driving and intelligent monitoring is semantic segmentation, and all pixel points in the images are subjected to target classification to obtain position information of environmental targets such as roads, pedestrians and automobiles, so that the automobiles are helped to avoid and run. However, the image semantic segmentation method generally uses a method of a full convolution network, such as a neural network, e.g., FCN, UNet, etc., which outputs a segmented image of the same size after inputting a road image, and classifies each pixel in the image with respect to different objects and backgrounds, although these networks combine deep information with shallow information, they do not fully utilize multi-scale information included in the output of different layers of the network, so that there are disadvantages in the recognition of some objects, for example, when a road environment changes, such as a tunnel appears in front, the recognition accuracy of the network for the objects is not ideal.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a tunnel identification model and a tunnel identification method based on multi-scale edge feature detection, so as to detect whether a tunnel exists in front or not according to a driving passage image captured by a vehicle in automatic driving and intelligent monitoring, thereby giving early warning in time.
In order to achieve the above object, the present invention provides a tunnel identification model based on multi-scale edge feature detection, including:
the data set construction module is used for constructing a data set for model training by utilizing images of various driving roads;
the training sample loading module is used for loading training samples in batches during training, cutting the image size into the same size during loading, and inputting the image size into the recognition network;
the method comprises the steps of identifying a network, constructing the identification network based on a residual error network Resnet34, using an edge binary image as a label when training the network, guiding the network to learn edge characteristics when learning image characteristics, and combining multi-scale edge characteristics to predict a tunnel more accurately.
Preferably, the data set construction module constructs a data set for model training by using a driving road image captured by a vehicle in automatic driving and intelligent monitoring as an input image, an artificially labeled image edge binary image and information of whether a tunnel exists or not as labels.
Preferably, the data in the data set takes the driving road image and the edge binary image and the tunnel label as a training sample.
Preferably, the identification network is based on the residual network Resnet34, and the final feature map of each block is output by removing the fully-connected layer, the initial 7 × 7 convolutional layer and the two pooling layers in the residual network Resnet34 and dividing the layers into 4 blocks according to the number of channels, and the 1 × 1 convolutional layer is used to convert the multi-layer feature maps into one image respectively.
Preferably, in addition to the output feature map of the first block, the output feature maps of the other three blocks are subjected to 1 × 1 convolution and then are subjected to up-sampling by being connected with an deconvolution layer, so that the image is converted into an image with the same size as the input image.
Preferably, all the images finally obtained by block are combined, and a single image is obtained by using 1 × 1 convolutional layer transformation, so that 5 images are output in the recognition network with the same size as the original image, and the 5 image outputs are calculated through corresponding true value images by using an error function.
Preferably, after the final combined image output layer, convolution is performed using two 3 × 3 convolutional layers, then a 2 × 2 maximum pooling layer is used, followed by two fully connected layers, resulting in outputs of {0, 1} representing the presence and absence of tunnels in the image, respectively.
Preferably, the recognition network has a total of 5 image outputs and one classification output, each using cross-entropy loss, the total loss function being the sum of all the loss functions.
In order to achieve the above object, the present invention further provides a tunnel identification method based on multi-scale edge feature detection, including the following steps:
step S1, constructing a data set for model training by using images of various driving roads;
step S2, loading training samples in batches, cutting the image size into the same size during loading, and inputting the image size into a recognition network;
step S3, constructing a loss function and constructing an SGD optimizer;
step S4, inputting training samples into the network in batches, calculating loss by using the loss function constructed in the step S3, and performing back propagation by using the optimizer constructed in the step S3 to update network parameters;
and S5, repeating the steps S2 to S4 for multiple times, and performing iterative optimization on the network until the training is finished to obtain a final model.
Preferably, in step S1, a data set for model training is constructed by using the image of the driving road captured by the vehicle in automatic driving and intelligent monitoring as an input image, the manually labeled image edge binary image, and the presence or absence of tunnel information as labels.
Compared with the prior art, the tunnel recognition model and method based on multi-scale edge feature detection construct a data set for model training by utilizing images of various driving roads, load training samples in batches during training, cut the image size to be equal during loading, input the same size into a recognition network, guide the network to mainly learn edge features during learning the image features by utilizing the recognition network to use edge binary images as labels during training the network, and simultaneously combine the multi-scale edge features, so that the tunnel is predicted more accurately and early warning is given in time.
Drawings
FIG. 1 is a system architecture diagram of a tunnel recognition model based on multi-scale edge feature detection according to the present invention;
FIG. 2 is an overall block diagram of an identification network in an embodiment of the invention;
FIG. 3 is a flowchart illustrating steps of a tunnel recognition method based on multi-scale edge feature detection according to the present invention;
FIG. 4 is a flow chart of an embodiment of the present invention.
Detailed Description
Other advantages and capabilities of the present invention will be readily apparent to those skilled in the art from the present disclosure by describing the embodiments of the present invention with specific embodiments thereof in conjunction with the accompanying drawings. The invention is capable of other and different embodiments and its several details are capable of modification in various other respects, all without departing from the spirit and scope of the present invention.
Fig. 1 is a system architecture diagram of a tunnel recognition model based on multi-scale edge feature detection according to the present invention. As shown in fig. 1, the tunnel recognition model based on multi-scale edge feature detection of the present invention includes:
a data set constructing module 101 for constructing a data set for model training using images of various driving roads.
In an embodiment of the present invention, the data set constructing module 101 constructs a data set for model training by using an image of a driving road captured by a vehicle in automatic driving and intelligent monitoring as an input image, an artificially labeled image edge binary image, and information of whether a tunnel exists or not as labels. That is, there are two types of images in the data set, one type is images of various driving roads, i.e. original shot images, some of which are tunneled and some of which are not tunneled, and the other type is edge images of the above-mentioned driving road images, i.e. label images, which are manually labeled binary images, besides these two types of images, each original shot image also has a tunnel label, i.e. whether the image is a tunnel or not is identified, which can be an integer or a character, e.g. 1 or not, which can be stored by using a text file alone, and the labels in the original shot image are in one-to-one correspondence with the input images and loaded during training. And taking the driving road image, the edge binary image and the tunnel label as a training sample. The constructed data set consists of three parts, namely a training set, a verification set and a test set.
The training sample loading module 102 is configured to load training samples in batch during training, that is, load a plurality of training samples each time, cut the image size to the same size during loading, and input the cut image size into the recognition network 103. It should be noted that the driving road image in each training sample and the corresponding artificially labeled binary image of the edge thereof are adjusted to the same size.
The identification network 103 is constructed based on the residual error network Resnet34, edge images are used as labels when the network is trained, namely the network is guided to mainly learn edge features when learning image features, and multi-scale edge features are combined, so that a tunnel can be predicted more accurately.
As shown in fig. 2, the main parts of the network 103 are identified as a residual network respet 34, the residual network respet 34 is divided into 4 blocks (each block is a processing in the residual network respet 34 including convolution, residual unit, etc.) according to the number of channels, the final feature map of each block is output, the feature maps of the multiple layers are converted into one image by using a convolutional layer of 1 × 1-1, and then the feature maps of the three blocks are down-sampled except the output feature map of the first block (not shown in the figure, generally, if the output image size of each block is based on the input image size, the block is 1-original size, the block is 2-2 times down-sampled, the block is 3-4 times down-sampled, and the block is 4-8 times down-sampled), further connecting an deconvolution layer (Deconv) to perform up-sampling, and converting the image into an image with the same size as the input image; combining the images finally obtained by all blocks, and obtaining an image by using convolution layer transformation of 1 x 1-1, so that 5 images are output in the same size as the original image in the network, the 5 image outputs are calculated by error functions through corresponding true value images (namely label images corresponding to input images), the true value images are edge images in a data set, and the loss function is cross entropy loss; after the finally combined image output layer, performing convolution by using two convolution layers of 3 × 3, then using a maximum pooling layer of 2 × 2, and then connecting two full-connection layers to obtain output {0, 1}, which respectively represents whether a tunnel exists or not in the image, wherein a true value is information on whether the tunnel exists or not in the label, and the loss function is a cross entropy loss function. Thus, the network uses cross entropy loss for each of a total of 6 outputs, 5 image outputs, and a classification output, with the total loss function being the sum of all the loss functions.
It should be noted that, in the training of the present invention, training samples are loaded from the training set for training, after each training round, the samples in the verification set are used for verification, so as to determine the training result, and after the training is completed successfully, the samples in the test set are used for testing and evaluating the model, and the processes of verification and testing are similar to those of training, which is not repeated herein.
Fig. 3 is a flowchart illustrating steps of a tunnel identification method based on multi-scale edge feature detection according to the present invention. As shown in fig. 3, the tunnel identification method based on multi-scale edge feature detection of the present invention includes the following steps:
in step S1, a data set for model training is constructed using images of various types of roads.
In the specific embodiment of the invention, a driving path image captured by a vehicle in automatic driving and intelligent monitoring is taken as input, an image edge binary image labeled manually and information of whether a tunnel exists or not are taken as labels, the input image, the edge image and the tunnel labels are taken as a training sample for data, and a data set comprises a training set, a verification set and a test set.
And step S2, loading training samples in batches, namely loading a plurality of training samples each time, cutting the image size into the same size during loading, and inputting the same size into the recognition network. It should be noted that the driving road image in each training sample and the corresponding artificially labeled binary image of the edge thereof are adjusted to the same size.
And step S3, constructing a loss function and constructing an SGD optimizer. In a specific embodiment of the invention, the 6 outputs of the network are combined with the corresponding labels to calculate the cross-entropy loss, which is then added as a loss function.
And step S4, inputting training samples into the network in batches, calculating loss by using the loss function constructed in the step S3, and performing back propagation by using the optimizer constructed in the step S3 to update network parameters.
And S5, repeating the steps S2 to S4 for multiple times, and performing iterative optimization on the network until the training is finished to obtain a final model.
Examples
As shown in fig. 4, in this embodiment, a tunnel identification method based on multi-scale edge feature detection includes the following specific implementation steps:
s1, making a data set, taking a driving path image captured by a vehicle in automatic driving and intelligent monitoring as input, taking an image edge binary image labeled manually and information of whether a tunnel exists as labels, taking the input image, the edge image and the tunnel labels as a training sample, and making the data set comprise a training set, a verification set and a test set.
And S2, loading data in batches, loading a plurality of samples at a time, and cutting the image size into the same size.
S3, constructing a loss function, calculating cross entropy loss by using the 6 outputs of the network and corresponding labels, and then adding the cross entropy loss as the loss function. In this embodiment, the loss function is cross entropy loss, and the formula is:
where, y is the tag value,is the network output result. Both the image output and the 0/1 output, whether predicted to have a tunnel or not, are calculated using this formula, resulting in a Loss of Loss. The specific implementation uses a pytorech framework in which a cross-entropy Loss function can directly compute Loss.
And S4, constructing the SGD optimizer. SGD is random gradient descent, after calculating the Loss value, the gradient value of each parameter weight in the network relative to the Loss is calculated, and then the network weight is adjusted according to the gradient value to realize the Loss value descent. A specific construction may use a pytore framework, where the gradient values of network weights with respect to Loss may be computed directly and adjusted.
And S5, inputting the input data in the samples into the network in batches, generating an output result by the network, calculating loss by using the loss function constructed in the S3, and then performing back propagation by using the optimizer constructed in the S4 to update network parameters.
And S6, repeating the steps 2 to 5 for multiple times, and carrying out iterative optimization on the network until the training is finished.
In summary, the tunnel recognition model and method based on multi-scale edge feature detection construct a data set for model training by using images of various driving roads, load training samples in batches during training, cut the image size to the same size during loading, input the same size into a recognition network, use an edge binary image as a label during network training by using the recognition network, guide the network to mainly learn edge features during image feature learning, and combine the multi-scale edge features, so that a tunnel is predicted more accurately and early warning is given in time.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Modifications and variations can be made to the above-described embodiments by those skilled in the art without departing from the spirit and scope of the present invention. Therefore, the scope of the invention should be determined from the following claims.
Claims (10)
1. A tunnel recognition model based on multi-scale edge feature detection comprises:
the data set construction module is used for constructing a data set for model training by utilizing images of various driving roads;
the training sample loading module is used for loading training samples in batches during training, cutting the image size into the same size during loading, and inputting the image size into the recognition network;
the method comprises the steps of identifying a network, constructing the identification network based on a residual error network Resnet34, using an edge binary image as a label when training the network, guiding the network to learn edge characteristics when learning image characteristics, and combining multi-scale edge characteristics to predict a tunnel more accurately.
2. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 1, wherein: the data set construction module constructs a data set for model training by using a driving road image captured by a vehicle in automatic driving and intelligent monitoring as an input image, and manually marked image edge binary images and information of whether a tunnel exists as labels.
3. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 2, wherein: and the data in the data set takes a driving road image, an edge binary image and a tunnel label as a training sample.
4. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 2, wherein: the identification network is based on a residual network Resnet34, a full connection layer, an initial 7 x 7 convolutional layer and two pooling layers in the residual network Resnet34 are removed, the two pooling layers are divided into 4 blocks according to the number of channels, the final feature map of each block is output, and the 1 x 1 convolutional layers are used for converting the multi-layer feature maps into one image respectively.
5. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 4, wherein: except the output characteristic diagram of the first block, the output characteristic diagrams of the other three blocks are subjected to convolution of 1 multiplied by 1 and then are connected with an deconvolution layer for up-sampling so as to transform the image into an image with the same size as the input image.
6. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 5, wherein: combining all the images finally obtained by block, and obtaining an image by using 1 x 1 convolutional layer transformation, so that the output of the identification network with the same size as the original image has 5 images, and the 5 image outputs are subjected to error function calculation through corresponding true value images.
7. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 6, wherein: after the final combined image output layer, convolution is performed using two convolution layers of 3 × 3, then a maximum pooling layer of 2 × 2 is used, and then two full-connected layers are connected to obtain outputs of {0, 1} representing the presence and absence of tunnels in the image, respectively.
8. The tunnel identification model based on multi-scale edge feature detection as claimed in claim 7, wherein: the recognition network has a total of 5 image outputs and one classification output, all using cross entropy loss, the total loss function being the sum of all loss functions.
9. A tunnel identification method based on multi-scale edge feature detection comprises the following steps:
step S1, constructing a data set for model training by using images of various driving roads;
step S2, loading training samples in batches, cutting the image size into the same size during loading, and inputting the image size into a recognition network;
step S3, constructing a loss function and constructing an SGD optimizer;
step S4, inputting training samples into the network in batches, calculating loss by using the loss function constructed in the step S3, and performing back propagation by using the optimizer constructed in the step S3 to update network parameters;
and S5, repeating the steps S2 to S4 for multiple times, and performing iterative optimization on the network until the training is finished to obtain a final model.
10. The tunnel identification method based on multi-scale edge feature detection as claimed in claim 9, wherein: in step S1, a data set for model training is constructed by using the image of the driving road captured by the vehicle in automatic driving and intelligent monitoring as an input image, the manually labeled image edge binary image, and the presence or absence of tunnel information as labels.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011007456.2A CN112085001B (en) | 2020-09-23 | Tunnel identification model and method based on multi-scale edge feature detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011007456.2A CN112085001B (en) | 2020-09-23 | Tunnel identification model and method based on multi-scale edge feature detection |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112085001A true CN112085001A (en) | 2020-12-15 |
CN112085001B CN112085001B (en) | 2024-04-23 |
Family
ID=
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112288044A (en) * | 2020-12-24 | 2021-01-29 | 成都索贝数码科技股份有限公司 | News picture attribute identification method of multi-scale residual error network based on tree structure |
CN115762629A (en) * | 2022-11-30 | 2023-03-07 | 天津大学 | Method for identifying interaction of enhancer and promoter |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180336683A1 (en) * | 2017-05-18 | 2018-11-22 | Mitsubishi Electric Research Laboratories, Inc. | Multi-Label Semantic Boundary Detection System |
CN109670404A (en) * | 2018-11-23 | 2019-04-23 | 江苏理工学院 | A kind of road ponding image detection method for early warning based on mixed model |
WO2019101221A1 (en) * | 2017-12-11 | 2019-05-31 | 珠海大横琴科技发展有限公司 | Ship detection method and system based on multidimensional scene characteristics |
CN110070008A (en) * | 2019-04-04 | 2019-07-30 | 中设设计集团股份有限公司 | Bridge disease identification method adopting unmanned aerial vehicle image |
CN110263660A (en) * | 2019-05-27 | 2019-09-20 | 魏运 | A kind of traffic target detection recognition method of adaptive scene changes |
CN110555857A (en) * | 2019-08-19 | 2019-12-10 | 浙江工业大学 | semantic edge dominant high-resolution remote sensing image segmentation method |
CN110796009A (en) * | 2019-09-29 | 2020-02-14 | 航天恒星科技有限公司 | Method and system for detecting marine vessel based on multi-scale convolution neural network model |
CN111368846A (en) * | 2020-03-19 | 2020-07-03 | 中国人民解放军国防科技大学 | Road ponding identification method based on boundary semantic segmentation |
CN111489297A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and apparatus for generating learning image data set for detecting dangerous elements |
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180336683A1 (en) * | 2017-05-18 | 2018-11-22 | Mitsubishi Electric Research Laboratories, Inc. | Multi-Label Semantic Boundary Detection System |
WO2019101221A1 (en) * | 2017-12-11 | 2019-05-31 | 珠海大横琴科技发展有限公司 | Ship detection method and system based on multidimensional scene characteristics |
CN109670404A (en) * | 2018-11-23 | 2019-04-23 | 江苏理工学院 | A kind of road ponding image detection method for early warning based on mixed model |
CN111489297A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and apparatus for generating learning image data set for detecting dangerous elements |
CN110070008A (en) * | 2019-04-04 | 2019-07-30 | 中设设计集团股份有限公司 | Bridge disease identification method adopting unmanned aerial vehicle image |
CN110263660A (en) * | 2019-05-27 | 2019-09-20 | 魏运 | A kind of traffic target detection recognition method of adaptive scene changes |
CN110555857A (en) * | 2019-08-19 | 2019-12-10 | 浙江工业大学 | semantic edge dominant high-resolution remote sensing image segmentation method |
CN110796009A (en) * | 2019-09-29 | 2020-02-14 | 航天恒星科技有限公司 | Method and system for detecting marine vessel based on multi-scale convolution neural network model |
CN111368846A (en) * | 2020-03-19 | 2020-07-03 | 中国人民解放军国防科技大学 | Road ponding identification method based on boundary semantic segmentation |
Non-Patent Citations (2)
Title |
---|
朱威;王图强;陈悦峰;何德峰;: "基于多尺度残差网络的对象级边缘检测算法", 计算机科学, no. 06 * |
胡珉;周显威;高新闻;: "公路隧道视频预处理和病害识别算法", 华侨大学学报(自然科学版), no. 05 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112288044A (en) * | 2020-12-24 | 2021-01-29 | 成都索贝数码科技股份有限公司 | News picture attribute identification method of multi-scale residual error network based on tree structure |
CN115762629A (en) * | 2022-11-30 | 2023-03-07 | 天津大学 | Method for identifying interaction of enhancer and promoter |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111507370B (en) | Method and device for obtaining sample image of inspection tag in automatic labeling image | |
JP6932395B2 (en) | A method for automatically evaluating the labeling reliability of a training image for use in a deep learning network for analyzing an image, and a reliability evaluation device using this method. | |
CN110197205B (en) | Image identification method of multi-feature-source residual error network | |
CN111598095A (en) | Deep learning-based urban road scene semantic segmentation method | |
CN110348384B (en) | Small target vehicle attribute identification method based on feature fusion | |
CN109886066A (en) | Fast target detection method based on the fusion of multiple dimensioned and multilayer feature | |
CN111950453A (en) | Optional-shape text recognition method based on selective attention mechanism | |
CN105574550A (en) | Vehicle identification method and device | |
CN111126134B (en) | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator | |
CN114495029B (en) | Traffic target detection method and system based on improved YOLOv4 | |
KR20200027889A (en) | Learning method, learning device for detecting lane using cnn and test method, test device using the same | |
CN111832615A (en) | Sample expansion method and system based on foreground and background feature fusion | |
CN111008626A (en) | Method and device for detecting object based on R-CNN | |
CN115937659A (en) | Mask-RCNN-based multi-target detection method in indoor complex environment | |
CN116597326A (en) | Unmanned aerial vehicle aerial photography small target detection method based on improved YOLOv7 algorithm | |
CN112364974A (en) | Improved YOLOv3 algorithm based on activation function | |
CN112215301B (en) | Image straight line detection method based on convolutional neural network | |
CN112800934A (en) | Behavior identification method and device for multi-class engineering vehicle | |
CN112380918A (en) | Road vehicle state identification method and device, electronic equipment and storage medium | |
CN112288702A (en) | Road image detection method based on Internet of vehicles | |
CN112085001B (en) | Tunnel identification model and method based on multi-scale edge feature detection | |
CN112085001A (en) | Tunnel recognition model and method based on multi-scale edge feature detection | |
CN113763447B (en) | Method for completing depth map, electronic device and storage medium | |
CN113989518A (en) | Lightweight target detection method applied to intelligent terminal | |
CN117456480B (en) | Light vehicle re-identification method based on multi-source information fusion |
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 |