CN109903323A - Training method, device, storage medium and terminal for transparent substance identification - Google Patents
Training method, device, storage medium and terminal for transparent substance identification Download PDFInfo
- Publication number
- CN109903323A CN109903323A CN201910167767.6A CN201910167767A CN109903323A CN 109903323 A CN109903323 A CN 109903323A CN 201910167767 A CN201910167767 A CN 201910167767A CN 109903323 A CN109903323 A CN 109903323A
- Authority
- CN
- China
- Prior art keywords
- depth
- image
- rgb
- images
- trained
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of training method, device, storage medium and terminals for transparent substance identification, method includes the following steps: S1, first data set of the foundation with multiple RGB images and the second data set with multiple depth images, multiple RGB images are corresponded with multiple depth images respectively；S2, the depth convolutional neural networks structure N1 for establishing multi-modal fusion；S3, multi-modal shared depth convolutional network structure N2 is established, N2 described in the fisrt feature information and the second feature information input is subjected to Fusion training, to export the sorting parameter information of object and location coordinate information and obtain network weight model M 2；S4, other multipair RGB images and depth image are re-entered parameter regulations is carried out to the network weight model M 1 and 2 pairs of the network weight model M with the network weight model M 11 and M22 after being optimized.
Description
Technical field
The present invention relates to image identification technical field, in particular to a kind of training method for transparent substance identification, dress
It sets, storage medium and terminal.
Background technique
Nowadays scientific and technological high speed development, the universal of industrial robot have not only liberated labour, have also speeded up speed of production, mentioned
The high quality of production.Wherein, the introducing of machine vision more makes robot grab improved efficiency.But for certain special
Article such as transparent substance, there is also accuracy of identification is not high or the difficulties such as time-consuming for machine vision.
Since the image of transparent substance is easy to be influenced by different factors etc., these factors affect to a certain extent
The stability and accuracy of single mode object identification system.Currently used method is that it is main to increase object by changing environment
Feature, but that this kind of methods are directed to and translucent etc object, such as patent CN104180772A, the requirement of identification
It is the rough surface of transparent substance, and can only identifies plate transparent substance.Meanwhile often these method equipment configratioin requirements are high, or
Person's calculating complexity is all difficult to meet industrial existence requirement, and such as patent CN102753933B, setting environment is harsh, needs to shield
External light source does not have practical application value industrially, can not adapt to changeable complex environment.
Therefore, the prior art is defective, needs to improve.
Summary of the invention
The embodiment of the present invention provides a kind of training method, device, storage medium and terminal for transparent substance identification, can
To improve the accuracy and efficiency of transparent substance identification.
The embodiment of the present invention provides a kind of training method for transparent substance identification, comprising the following steps:
S1, the first data set with multiple RGB images and the second data set with multiple depth images are established, this multiple
RGB image is corresponded with multiple depth images respectively；
S2, depth convolutional neural networks the structure N1, the N1 for establishing multi-modal fusion are carried out individually for extracting multiple RGB images
It trains and extracts multiple depth images and individually trained, to extract the fisrt feature information and depth of RGB image respectively
The second feature information of image and obtain network weight model M 1；
S3, multi-modal shared depth convolutional network structure N2 is established, by the fisrt feature information and the second feature
N2 described in information input carries out Fusion training, to export the sorting parameter information of object and location coordinate information and obtain network
Weight model M2；
S4, other multipair RGB images and depth image are re-entered to the network weight model M 1 and the network weight
2 pairs of model M carry out parameter regulations with the network weight model M 11 and M22 after being optimized.
In the training method of the present invention for transparent substance identification, described establish has multiple RGB images
First data set and the second data set with multiple depth images, multiple RGB images respectively with multiple depth images
The step of one-to-one correspondence includes:
Acquire the RGB image and depth image of multiple objects to be trained, multiple RGB images respectively with multiple depth images
It corresponds；
Boundary demarcation is carried out to the object to be trained in the RGB image, and the first classification ginseng of the object to be trained is set
The first position coordinate information of number information and the object to be trained in the RGB image；
First number with multiple RGB images is established according to the first sorting parameter information, the first position coordinate information
According to collection；
Side is carried out to the object to be trained in the depth image according to the corresponding relationship of the RGB image and the depth image
Boundary mark is fixed, and the second sorting parameter information that the object to be trained is arranged and the object to be trained are in the depth image
In second position coordinate information；
Second number with multiple depth images is established according to the second sorting parameter information, the second position coordinate information
According to collection.
In the training method of the present invention for transparent substance identification, the depth volume for establishing multi-modal fusion
Product neural network structure N1, the N1 carry out individually training and extracting the progress of multiple depth images for extracting multiple RGB images
Individually training, with extract respectively RGB image fisrt feature information and depth image second feature information and obtain network
The step of weight model M1 includes:
Depth convolutional neural networks the structure N1, the N1 for establishing multi-modal fusion include two independent convolutional neural networks point
Branch, this two independent convolutional neural networks branches are for being individually trained the RGB image and depth image；Its
In, in training, mutual corresponding RGB image and depth image are randomly selected from the first data set and the second data set every time
As input, the fisrt feature information of RGB image and the second feature of depth image are extracted respectively using convolutional neural networks
Information and obtain network weight model M 1.
In the training method of the present invention for transparent substance identification, the mutual corresponding RGB image and depth
Degree image is the image that the same object of colored RGB camera and depth camera acquisition is respectively adopted.
In the training method of the present invention for transparent substance identification, in the step S2, returned using reversed
Propagation algorithm and the error by returning loss layer update each layer of parameter, so that network weight model is able to update optimization,
Final convergence.
A kind of training device for transparent substance identification, comprising:
First establishes module, for establishing the first data set with multiple RGB images and with multiple depth images
Two data sets, multiple RGB images are corresponded with multiple depth images respectively；
Second establishes module, for establishing depth convolutional neural networks the structure N1, the N1 of multi-modal fusion for extracting multiple
RGB image individually train and extract multiple depth images individually being trained, special with extract RGB image respectively first
Reference breath and depth image second feature information and obtain network weight model M 1；
Third establishes module, for establishing multi-modal shared depth convolutional network structure N2, by the fisrt feature information with
And N2 described in the second feature information input carries out Fusion training, to export the sorting parameter information and position coordinates of object
Information simultaneously obtains network weight model M 2；
Optimization module, for re-entering other multipair RGB images and depth image to the network weight model M 1 and institute
It states 2 pairs of network weight model M and carries out parameter regulations with the network weight model M 11 and M22 after being optimized.
In the training device of the present invention for transparent substance identification, described first, which establishes module, includes:
Acquisition unit, for acquiring the RGB image and depth image of multiple objects to be trained, multiple RGB images respectively with
Multiple depth images correspond；
First calibration unit for carrying out boundary demarcation to the object to be trained in the RGB image, and is arranged described wait train
First position coordinate information of the first sorting parameter information and the object to be trained of object in the RGB image；
First establishing unit, for being established according to the first sorting parameter information, the first position coordinate information with more
Open the first data set of RGB image；
Second calibration unit, for the corresponding relationship according to the RGB image and the depth image in the depth image
Object to be trained carry out boundary demarcation, and the second sorting parameter information of the object to be trained and described wait train is set
Second position coordinate information of the object in the depth image；
Second establishes unit, for being established according to the second sorting parameter information, the second position coordinate information with more
Open the second data set of depth image.
In the training device of the present invention for transparent substance identification, the mutual corresponding RGB image and depth
Degree image is the image that the same object of colored RGB camera and depth camera acquisition is respectively adopted.
A kind of storage medium is stored with computer program in the storage medium, when the computer program is in computer
When upper operation, so that the computer executes method described in any of the above embodiments.
A kind of terminal, including processor and memory, computer program is stored in the memory, and the processor is logical
The computer program for calling and storing in the memory is crossed, for executing method described in any of the above embodiments.
The present invention passes through a series of nerves by allowing the data (RGB image and depth image) of different modalities individually to train
Network, study arrive the feature of mode itself, are then got up by fusion connection, by a series of shared convolutional layers, to each mode
Feature carry out complementary study, the fusion of RGB information and depth information, which can achieve, promotes the effect that transparent substance identifies.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described.It should be evident that drawings in the following description are only some embodiments of the invention, for
For those skilled in the art, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is the flow diagram of the training method provided in an embodiment of the present invention for transparent substance identification.
Fig. 2 is the structural schematic diagram of the training device provided in an embodiment of the present invention for transparent substance identification.
Fig. 3 is the structural schematic diagram of terminal provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those skilled in the art's every other implementation obtained under that premise of not paying creative labor
Example, belongs to protection scope of the present invention.
Description and claims of this specification and term " first " in above-mentioned attached drawing, " second ", " third " etc.
(if present) is to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be appreciated that this
The object of sample description is interchangeable under appropriate circumstances.In addition, term " includes " and " having " and their any deformation, meaning
Figure, which is to cover, non-exclusive includes.For example, containing the process, method of series of steps or containing a series of modules or list
The device of member, advanced DAS (Driver Assistant System), system those of are not necessarily limited to be clearly listed step or module or unit, can be with
Include the steps that being not clearly listed or module or unit, also may include for these process, methods, device, advanced auxiliary
The other steps or module or unit for helping control loop or system intrinsic.
With reference to Fig. 1, Fig. 1 is a kind of flow chart of training method for transparent substance identification.This is used for transparent substance knowledge
Other training method, comprising the following steps:
S1, the first data set with multiple RGB images and the second data set with multiple depth images are established, this multiple
RGB image is corresponded with multiple depth images respectively.
Wherein, use the picture to training image that shoots in real scene as training sample.
Specifically, step S1 includes:
S11, the RGB image and depth image for acquiring multiple objects to be trained, multiple RGB images respectively with multiple depth
Image corresponds；S12, boundary demarcation is carried out to the object to be trained in the RGB image, and the object to be trained is set
First position coordinate information in the RGB image of the first sorting parameter information and the object to be trained；S13, root
First data set with multiple RGB images is established according to the first sorting parameter information, the first position coordinate information；
S14, side is carried out to the object to be trained in the depth image according to the corresponding relationship of the RGB image and the depth image
Boundary mark is fixed, and the second sorting parameter information that the object to be trained is arranged and the object to be trained are in the depth image
In second position coordinate information；S15, tool is established according to the second sorting parameter information, the second position coordinate information
There is the second data set of multiple depth images.
Wherein, mutual corresponding RGB image and depth image are that colored RGB camera is respectively adopted to acquire with depth camera
The image of the same object.
In step sl, colored RGB block and depth image module are in the different location of sensor.So even
In the same time to same object, acquired image information is all different.Since we need to unitize to bounding box,
So needing color RGB image and deep image information carrying out matrixing, correspond its coordinate.Here, it needs to examine
Consider two matrixes, one is translation matrix, and one is spin matrix.
It is assumed that depth image certain point is (x, y), corresponding point is that (X, Y) therefore passes through on color RGB image
Translation matrix, we are available X=x+dx, and Y=y+dy is expressed as follows with matrix.
Dx and dy is x, the distance that y is moved in the direction respectively.Translation matrix is expressed as follows:
Meanwhile there are one spin matrixs, if certain point and origin line and X-axis angle are b degree, using origin as the center of circle, counterclockwise
A degree is turned over, origin and the wire length are R, and [x, y] is depth image coordinate, and [X, Y] is color RGB image coordinate, because
This, can obtain:
；
；
。
It is consequently possible to calculate it is as follows to obtain spin matrix:
。
S2, depth convolutional neural networks the structure N1, the N1 of multi-modal fusion are established for extracting the progress of multiple RGB images
Individually training and extract multiple depth images and individually trained, with extract respectively RGB image fisrt feature information and
The second feature information of depth image and obtain network weight model M 1.
Wherein, step S2 includes: the depth convolutional neural networks structure N1 for establishing multi-modal fusion, and the N1 includes two
A independent convolutional neural networks branch, this two independent convolutional neural networks branches are for individually scheming the RGB
Picture and depth image are trained；Wherein, it in training, is randomly selected from the first data set and the second data set every time mutually
Corresponding RGB image and depth image extract the fisrt feature information of RGB image using convolutional neural networks as input respectively
And depth image second feature information and obtain network weight model M 1.
Wherein it is possible to update each layer of parameter using reversed passback algorithm and the error by returning loss layer, make
Network weight model is obtained to be able to update optimization, it is final to restrain.
S3, multi-modal shared depth convolutional network structure N2 is established, by the fisrt feature information and described second
Characteristic information inputs the N2 and carries out Fusion training, to export the sorting parameter information of object and location coordinate information and obtain
Network weight model M 2.
Wherein, which includes multiple convolutional neural networks, then connects multiple fully-connected networks, and output includes two ginsengs
Number, one be object coordinate position parameter, one be object sorting parameter.
S4, other multipair RGB images and depth image are re-entered to the network weight model M 1 and the network
Weight model M2 is to carrying out parameter regulation with the network weight model M 11 and M22 after being optimized.
Using trained network weight model M 1 and M2, new data input network, weight are extracted from data set again
Small parameter perturbations newly are carried out to global network, realize the hiding relationship for finding input and output.It is separately trained by two parts network,
Training time cost can be reduced.
In this application, the network structure of N1 and N2, including it is singly not limited to convolutional layer, pond layer, nonlinear function layer, entirely
Articulamentum normalizes layer, and includes but is not limited to any combination of these layers, and the structure of network is not the range of protection.
Referring to figure 2., a kind of training device for transparent substance identification, comprising: first establishes module 201, second builds
Formwork erection block 202, third establish module 203 and optimization module 204.
Wherein, this first establishes module 201 for establishing the first data set with multiple RGB images and with multiple
Second data set of depth image, multiple RGB images are corresponded with multiple depth images respectively.Wherein, it corresponds to each other
RGB image and depth image be respectively adopted colored RGB camera and depth camera acquisition the same object image.
Wherein, this first to establish module include: acquisition unit, for acquire multiple objects to be trained RGB image and
Depth image, multiple RGB images are corresponded with multiple depth images respectively；First calibration unit, for the RGB
Object to be trained in image carries out boundary demarcation, and the first sorting parameter information of the object to be trained and described is arranged
First position coordinate information of the object to be trained in the RGB image；First establishing unit, for according to first classification
Parameter information, the first position coordinate information establish first data set with multiple RGB images；Second calibration unit, is used
In the corresponding relationship according to the RGB image and the depth image to object the to be trained progress boundary in the depth image
Calibration, and the second sorting parameter information that the object to be trained is arranged and the object to be trained are in the depth image
Second position coordinate information；Second establishes unit, for according to the second sorting parameter information, the second position coordinate
Information establishes second data set with multiple depth images.
Wherein, this second establishes module 202 for establishing depth convolutional neural networks the structure N1, the N1 of multi-modal fusion
Individually train and extract multiple depth images individually being trained for extracting multiple RGB images, to extract RGB respectively
The fisrt feature information of image and the second feature information of depth image and obtain network weight model M 1.Wherein, this second
Establishing module 202 and establishing depth convolutional neural networks the structure N1, the N1 of multi-modal fusion includes two independent convolution minds
Through network branches, this two independent convolutional neural networks branches for individually to the RGB image and depth image into
Row training；Wherein, in training, mutual corresponding RGB image is randomly selected from the first data set and the second data set every time
With depth image as inputting, the fisrt feature information and depth image of RGB image are extracted respectively using convolutional neural networks
Second feature information and obtain network weight model M 1.
Wherein, which establishes module 203 for establishing multi-modal shared depth convolutional network structure N2, by described the
N2 described in one characteristic information and the second feature information input carries out Fusion training, to export the sorting parameter information of object
And location coordinate information and obtain network weight model M 2.Wherein, which includes multiple convolutional neural networks, is then connected more
A fully-connected network, output include two parameters, one be object coordinate position parameter, one be object sorting parameter.
Wherein, the optimization module 204 is for re-entering other multipair RGB images and depth image to the network weight
Model M 1 and 2 pairs of the network weight model M carry out parameter regulations with after being optimized network weight model M 11 and
M22。
Using trained network weight model M 1 and M2, new data input network, weight are extracted from data set again
Small parameter perturbations newly are carried out to global network, realize the hiding relationship for finding input and output.It is separately trained by two parts network,
Training time cost can be reduced.
Finally, after being optimized network weight model M 11 and M22 after, can use the network weight model M 11
And M22 identifies transparent substance, and has high accuracy and high efficiency.
The present invention also provides a kind of storage medium, it is stored with computer program in the storage medium, when the calculating
When machine program is run on computers, identified described in any of the above-described embodiment for transparent substance so that the computer executes
Training.
Referring to figure 3., the present invention also provides a kind of terminal, terminal includes processor 301 and memory 302.Wherein, locate
It manages device 301 and memory 302 is electrically connected.
Processor 301 is the control centre of terminal, using the various pieces of various interfaces and the entire terminal of connection, is led to
It crosses operation or calls the computer program being stored in memory 302, and call the data being stored in memory 302, hold
The various functions and processing data of row terminal, to carry out integral monitoring to terminal.
In the present embodiment, the processor 301 in terminal can be according to following step, by one or more calculating
The corresponding instruction of the process of machine program is loaded into memory 302, and runs storage in the memory 302 by processor 301
Computer program, to realize various functions: establishing the first data set with multiple RGB images and with multiple depth
Second data set of image, multiple RGB images are corresponded with multiple depth images respectively；Establish the depth of multi-modal fusion
Convolutional neural networks structure N1 is spent, which carries out individually training and extracting multiple depth images for extracting multiple RGB images
Individually trained, with extract respectively RGB image fisrt feature information and depth image second feature information and obtain
Network weight model M 1；Multi-modal shared depth convolutional network structure N2 is established, by the fisrt feature information and described
N2 described in second feature information input carry out Fusion training, with export object sorting parameter information and location coordinate information simultaneously
Obtain network weight model M 2；Re-enter other multipair RGB images and depth image to the network weight model M 1 and
2 pairs of the network weight model M carry out parameter regulations with the network weight model M 11 and M22 after being optimized.
It should be noted that those of ordinary skill in the art will appreciate that whole in the various methods of above-described embodiment or
Part steps are relevant hardware can be instructed to complete by program, which can store in computer-readable storage medium
In matter, which be can include but is not limited to: read-only memory (ROM, Read Only Memory), random access memory
Device (RAM, Random Access Memory), disk or CD etc..
Be provided for the embodiments of the invention above based on it is advanced drive auxiliary based reminding method, device, storage medium and
Advanced DAS (Driver Assistant System) is described in detail, specific case used herein to the principle of the present invention and embodiment into
Elaboration is gone, the above description of the embodiment is only used to help understand the method for the present invention and its core ideas；Meanwhile for this
The technical staff in field, according to the thought of the present invention, there will be changes in the specific implementation manner and application range, to sum up
Described, the contents of this specification are not to be construed as limiting the invention.
Claims (10)
1. a kind of training method for transparent substance identification, which comprises the following steps:
S1, the first data set with multiple RGB images and the second data set with multiple depth images are established, this multiple
RGB image is corresponded with multiple depth images respectively；
S2, depth convolutional neural networks the structure N1, the N1 for establishing multi-modal fusion are carried out individually for extracting multiple RGB images
It trains and extracts multiple depth images and individually trained, to extract the fisrt feature information and depth of RGB image respectively
The second feature information of image and obtain network weight model M 1；
S3, multi-modal shared depth convolutional network structure N2 is established, by the fisrt feature information and the second feature
N2 described in information input carries out Fusion training, to export the sorting parameter information of object and location coordinate information and obtain network
Weight model M2；
S4, other multipair RGB images and depth image are re-entered to the network weight model M 1 and the network weight
2 pairs of model M carry out parameter regulations with the network weight model M 11 and M22 after being optimized.
2. the training method according to claim 1 for transparent substance identification, which is characterized in that described to establish with more
It opens the first data set of RGB image and the second data set with multiple depth images, multiple RGB images is more with this respectively
Opening the step of depth image corresponds includes:
Acquire the RGB image and depth image of multiple objects to be trained, multiple RGB images respectively with multiple depth images
It corresponds；
Boundary demarcation is carried out to the object to be trained in the RGB image, and the first classification ginseng of the object to be trained is set
The first position coordinate information of number information and the object to be trained in the RGB image；
First number with multiple RGB images is established according to the first sorting parameter information, the first position coordinate information
According to collection；
Side is carried out to the object to be trained in the depth image according to the corresponding relationship of the RGB image and the depth image
Boundary mark is fixed, and the second sorting parameter information that the object to be trained is arranged and the object to be trained are in the depth image
In second position coordinate information；
Second number with multiple depth images is established according to the second sorting parameter information, the second position coordinate information
According to collection.
3. it is according to claim 1 for transparent substance identification training method, which is characterized in that it is described establish it is multi-modal
Depth convolutional neural networks the structure N1, the N1 of fusion individually train and extract multiple for extracting multiple RGB images
Depth image is individually trained, to extract the fisrt feature information of RGB image and the second feature letter of depth image respectively
Ceasing the step of obtaining network weight model M 1 includes:
Depth convolutional neural networks the structure N1, the N1 for establishing multi-modal fusion include two independent convolutional neural networks point
Branch, this two independent convolutional neural networks branches are for being individually trained the RGB image and depth image；Its
In, in training, mutual corresponding RGB image and depth image are randomly selected from the first data set and the second data set every time
As input, the fisrt feature information of RGB image and the second feature of depth image are extracted respectively using convolutional neural networks
Information and obtain network weight model M 1.
4. the training method according to claim 1 for transparent substance identification, which is characterized in that described mutual corresponding
RGB image and depth image are the image that the same object of colored RGB camera and depth camera acquisition is respectively adopted.
5. the training method according to claim 1 for transparent substance identification, which is characterized in that in the step S2
In, each layer of parameter is updated using reversed passback algorithm and the error by returning loss layer, so that network weight model
It is able to update optimization, it is final to restrain.
6. a kind of training device for transparent substance identification characterized by comprising
First establishes module, for establishing the first data set with multiple RGB images and with multiple depth images
Two data sets, multiple RGB images are corresponded with multiple depth images respectively；
Second establishes module, for establishing depth convolutional neural networks the structure N1, the N1 of multi-modal fusion for extracting multiple
RGB image individually train and extract multiple depth images individually being trained, special with extract RGB image respectively first
Reference breath and depth image second feature information and obtain network weight model M 1；
Third establishes module, for establishing multi-modal shared depth convolutional network structure N2, by the fisrt feature information with
And N2 described in the second feature information input carries out Fusion training, to export the sorting parameter information and position coordinates of object
Information simultaneously obtains network weight model M 2；
Optimization module, for re-entering other multipair RGB images and depth image to the network weight model M 1 and institute
It states 2 pairs of network weight model M and carries out parameter regulations with the network weight model M 11 and M22 after being optimized.
7. the training device according to claim 6 for transparent substance identification, which is characterized in that described first establishes mould
Block includes:
Acquisition unit, for acquiring the RGB image and depth image of multiple objects to be trained, multiple RGB images respectively with
Multiple depth images correspond；
First calibration unit for carrying out boundary demarcation to the object to be trained in the RGB image, and is arranged described wait train
First position coordinate information of the first sorting parameter information and the object to be trained of object in the RGB image；
First establishing unit, for being established according to the first sorting parameter information, the first position coordinate information with more
Open the first data set of RGB image；
Second calibration unit, for the corresponding relationship according to the RGB image and the depth image in the depth image
Object to be trained carry out boundary demarcation, and the second sorting parameter information of the object to be trained and described wait train is set
Second position coordinate information of the object in the depth image；
Second establishes unit, for being established according to the second sorting parameter information, the second position coordinate information with more
Open the second data set of depth image.
8. the training device according to claim 6 for transparent substance identification, which is characterized in that described mutual corresponding
RGB image and depth image are the image that the same object of colored RGB camera and depth camera acquisition is respectively adopted.
9. a kind of storage medium, which is characterized in that computer program is stored in the storage medium, when the computer program
When running on computers, so that the computer perform claim requires 1 to 5 described in any item methods.
10. a kind of terminal, which is characterized in that including processor and memory, computer program, institute are stored in the memory
Processor is stated by calling the computer program stored in the memory, requires any one of 1 to 5 institute for perform claim
The method stated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910167767.6A CN109903323B (en) | 2019-03-06 | 2019-03-06 | Training method and device for transparent object recognition, storage medium and terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910167767.6A CN109903323B (en) | 2019-03-06 | 2019-03-06 | Training method and device for transparent object recognition, storage medium and terminal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903323A true CN109903323A (en) | 2019-06-18 |
| CN109903323B CN109903323B (en) | 2022-11-18 |
Family
ID=66946615
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910167767.6A Active CN109903323B (en) | 2019-03-06 | 2019-03-06 | Training method and device for transparent object recognition, storage medium and terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903323B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458828A (en) * | 2019-08-12 | 2019-11-15 | 广东工业大学 | A kind of laser welding defect identification method and device based on multi-modal fusion network |
| CN112082475A (en) * | 2020-08-25 | 2020-12-15 | 中国科学院空天信息创新研究院 | Living tree species identification method and volume measurement method |
| CN116665002A (en) * | 2023-06-28 | 2023-08-29 | 北京百度网讯科技有限公司 | Image processing method, training method and device for deep learning model |
| CN117115208A (en) * | 2023-10-20 | 2023-11-24 | 城云科技（中国）有限公司 | Transparent object tracking model, construction method and application thereof |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108182441A (en) * | 2017-12-29 | 2018-06-19 | 华中科技大学 | Parallel multichannel convolutive neural network, construction method and image characteristic extracting method |
| US20180330194A1 (en) * | 2017-05-15 | 2018-11-15 | Siemens Aktiengesellschaft | Training an rgb-d classifier with only depth data and privileged information |
-
2019
- 2019-03-06 CN CN201910167767.6A patent/CN109903323B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180330194A1 (en) * | 2017-05-15 | 2018-11-15 | Siemens Aktiengesellschaft | Training an rgb-d classifier with only depth data and privileged information |
| CN108182441A (en) * | 2017-12-29 | 2018-06-19 | 华中科技大学 | Parallel multichannel convolutive neural network, construction method and image characteristic extracting method |
Non-Patent Citations (1)
| Title |
|---|
| 黄斌等: "基于深度卷积神经网络的物体识别算法", 《计算机应用》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458828A (en) * | 2019-08-12 | 2019-11-15 | 广东工业大学 | A kind of laser welding defect identification method and device based on multi-modal fusion network |
| CN110458828B (en) * | 2019-08-12 | 2023-02-10 | 广东工业大学 | Laser welding defect identification method and device based on multi-mode fusion network |
| CN112082475A (en) * | 2020-08-25 | 2020-12-15 | 中国科学院空天信息创新研究院 | Living tree species identification method and volume measurement method |
| CN112082475B (en) * | 2020-08-25 | 2022-05-24 | 中国科学院空天信息创新研究院 | Living stumpage species identification method and volume measurement method |
| CN116665002A (en) * | 2023-06-28 | 2023-08-29 | 北京百度网讯科技有限公司 | Image processing method, training method and device for deep learning model |
| CN116665002B (en) * | 2023-06-28 | 2024-02-27 | 北京百度网讯科技有限公司 | Image processing method, training method and device for deep learning model |
| CN117115208A (en) * | 2023-10-20 | 2023-11-24 | 城云科技（中国）有限公司 | Transparent object tracking model, construction method and application thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903323B (en) | 2022-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109903323A (en) | Training method, device, storage medium and terminal for transparent substance identification | |
| CN109325504A (en) | A kind of underwater sea cucumber recognition methods and system | |
| CN110020620A (en) | Face identification method, device and equipment under a kind of big posture | |
| Li et al. | A deep learning approach for real-time rebar counting on the construction site based on YOLOv3 detector | |
| CN109816725A (en) | A kind of monocular camera object pose estimation method and device based on deep learning | |
| CN110232716A (en) | A kind of camera calibration method, apparatus and electronic equipment | |
| CN110991435A (en) | Express waybill key information positioning method and device based on deep learning | |
| CN113538574B (en) | Pose positioning method, device and equipment and computer readable storage medium | |
| CN109461184B (en) | Automatic positioning method for grabbing point for grabbing object by robot mechanical arm | |
| CN109548489A (en) | A kind of apple picking robot based on yolo vision | |
| CN111524184B (en) | Intelligent unstacking method and unstacking system based on 3D vision | |
| CN114821014B (en) | Multi-mode and countermeasure learning-based multi-task target detection and identification method and device | |
| CN111931764A (en) | Target detection method, target detection framework and related equipment | |
| CN114549507B (en) | Improved Scaled-YOLOv fabric flaw detection method | |
| CN108090486A (en) | Image processing method and device in a kind of game of billiards | |
| CN115330734A (en) | Automatic robot repair welding system based on three-dimensional target detection and point cloud defect completion | |
| CN110142765A (en) | A kind of method, apparatus and system of processing rubber plug | |
| CN107527368A (en) | Three-dimensional attitude localization method and device based on Quick Response Code | |
| CN113657551A (en) | Robot grabbing posture task planning method for sorting and stacking multiple targets | |
| CN114131603B (en) | Deep reinforcement learning robot grabbing method based on perception enhancement and scene migration | |
| CN113838158B (en) | Image and video reconstruction method and device, terminal equipment and storage medium | |
| CN109760067B (en) | Intelligent robot system and equipment capable of playing cards | |
| CN110209860A (en) | A kind of interpretable garment coordination method and device based on clothes attribute of template-directed | |
| CN109089076A (en) | A kind of robot motion real-time monitoring system | |
| Aziz et al. | Evaluation of visual attention models for robots |
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 |