CN109785333A - Object detection method and device for parallel manipulator human visual system - Google Patents
Object detection method and device for parallel manipulator human visual system Download PDFInfo
- Publication number
- CN109785333A CN109785333A CN201811509164.1A CN201811509164A CN109785333A CN 109785333 A CN109785333 A CN 109785333A CN 201811509164 A CN201811509164 A CN 201811509164A CN 109785333 A CN109785333 A CN 109785333A
- Authority
- CN
- China
- Prior art keywords
- image
- data
- training
- target
- submodule
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to technical field of image processing, in particular to a kind of object detection method and device for parallel manipulator human visual system, this method includes: acquisition target image, and is divided into training dataset and test data set；Mixing autocoder is built, concentrates image to be sent into mixing autocoder in batches training data and carries out denoising and sparse processing, obtain the image data for having tag along sort；Image data is sent into Faster RCNN neural network and carries out model training；Test data set image data is sent into trained Faster RCNN neural network and carries out model measurement；By in the Faster RCNN neural network model insertion parallel manipulator human visual system after test, object recognition and detection is carried out to image to be detected.The present invention improves model anti-interference ability and its robustness, reduce data set training difficulty, model training speed and quality is improved, makes neural network be convenient for being easy to reuse for different targets to be identified, there is great importance to image procossing and target identification technology field.
Description
Technical field
The invention belongs to technical field of image processing, in particular to a kind of target for parallel manipulator human visual system is examined
Survey method and device.
Background technique
Machine vision development in recent years is swift and violent, and machine vision will more and more be applied to industry, life, traffic, boat
The industries scene such as empty space flight, amusement, construction becomes the eyes of artificial intelligence, therefore more fast and accurate machine vision will be
Later development trend, in technological layer, faster more accurately deep learning will be later research emphasis, by reducing model
The optimization machine vision such as training difficulty, increase model training recognition speed, will be the development trends of machine vision.In machine vision
In, target detection technique is applied to the fields such as all types of industries production, communications and transportation, plays indispensable role.Wherein
The target detection performance based on deep learning such as YOLO, R-CNN, Fast R-CNN, Faster R-CNN is good, but most of
Model spends big, the low problem of the slow precision of speed, therefore there are also many aspects to need for target detection there is also data set production is difficult
It improves.
Summary of the invention
For this purpose, the present invention provides a kind of object detection method and device for parallel manipulator human visual system, number is reduced
According to collection manufacture difficulty, model generalization ability is improved, further solution color is close, has the target object for blocking incompleteness detection to ask
Topic improves the speed of service and object recognition and detection precision.
According to design scheme provided by the present invention, a kind of object detection method for parallel manipulator human visual system,
Include following content:
Target image is acquired, and collected target image is divided into training dataset and test data set.By training set
In the picture comprising target and pure background picture by Background difference one accurately target candidate frame of acquisition, according to candidate frame pair
Image is sheared, and folder name is referred to as tag along sort in conjunction with the target that shearing obtains, and obtains the data for having label
Collection.Each group of classification obtains the data set of label all in accordance with this mode；
Build mixing autocoder, by training data concentrate image in batches is sent into mixing autocoder carry out denoising with
Sparse processing obtains the image of removal noise and sparse features；
Faster RCNN neural network will be sent into tag along sort image data carry out model training；
Two groups of target images to be detected that test data is concentrated are sent into trained Faster RCNN neural network
Carry out model measurement；
By in the Faster RCNN neural network model insertion parallel manipulator human visual system after test, taken the photograph by industry
Camera carries out object recognition and detection to image to be detected collected in field range.
Above-mentioned, it acquires in target image, which is divided into two groups, wherein one group is every figure
Piece only includes an object to be identified, and another group is there are a variety of difference objects to be identified in every picture；And it is to be checked by every group
It surveys target image and is divided into training data and test data, every group of corresponding image data is inputted into training dataset respectively
And test data set.
Above-mentioned, the mixing autocoder includes first layer denoising self-encoding encoder, second layer denoising self-encoding encoder, the
Three layers of sparse self-encoding encoder and monitor layer set on each encoder output；Using the output characteristics of image of three layer coders as
Corresponding every layer coder monitor layer input, carries out tag along sort processing to image data using monitor layer, obtain removal noise and
The image of sparse features.
Above-mentioned, the monitor layer classifies automatically to output characteristics of image and adds tag along sort, obtains to have and divide
The image data of class label.
It is above-mentioned, for the image data of input, Faster RCNN neural network by it is preceding to, back-propagation algorithm into
The test of row model training.
Above-mentioned, it obtains in alternative initial population process, introduces mechanism of chaos, and obtain alternative initial by tent maps
A element of volume in population.
A kind of object detecting device for parallel manipulator human visual system includes: acquisition module, coding module, training
Module, test module and target identification module, wherein
Acquisition module is divided into training dataset and test number for acquiring target image, and by collected target image
According to collection；
Coding module, for building mixing autocoder, concentrating image to be sent into batches training data, mixing is automatic to be compiled
Code device carries out denoising and sparse processing, obtains the image feature data for having tag along sort；
Training module is sent into Faster RCNN mind with tag along sort image feature data for export monitor layer
Model training is carried out through network；
Test module, two groups of target images to be detected for concentrating test data are sent into trained Faster
Model measurement is carried out in RCNN neural network；
Target identification module is embedded in parallel robot vision for the Faster RCNN neural network model after testing
In system, object recognition and detection is carried out to image to be detected collected in field range by industrial camera.
In above-mentioned device, acquisition module includes: acquisition submodule and grouping submodule, wherein
Submodule is acquired, collects target data for shooting, which includes that every picture only includes one wait know
There are the multi-Target Image data of a variety of objects Bu Tong to be identified in every picture for the single goal image data of other object；
It is grouped submodule, for the target data being collected into be divided into two containing physical quantities to be identified according in picture
Group, wherein one group is single goal image data set, and another group is multi-Target Image data group；And by every group of target figure to be detected
As being divided into training data and test data, every group of corresponding image data is inputted into training dataset and test number respectively
According to collection.
In above-mentioned device, the coding module include first denoising encoding submodule, second denoising encoding submodule,
Sparse coding submodule and supervised classification submodule, wherein
First denoising encoding submodule, for being that image data adds man-made noise, construction using denoising autocoder
Image training characteristics；
Second denoising encoding submodule, for being that image data adds man-made noise, construction using denoising autocoder
Image training characteristics；
Sparse coding submodule obtains output characteristics of image for carrying out image real time transfer using sparse self-encoding encoder；
Supervised classification submodule, for by denoising both autocoder and sparse self-encoding encoder output end addition prison
Layer is superintended and directed, the image feature data eventually with tag along sort is obtained.
In above-mentioned device, the supervised classification submodule includes data categorization module and addition label model, wherein
Data categorization module, for carrying out data classification to output characteristics of image；
Label model is added, for adding label automatically to sorted image feature data, is obtained with tag along sort
Image feature data.
Beneficial effects of the present invention:
The present invention solves traditional detection method data set and makes heavy workload, computationally intensive, or mesh close for color
The problem of accuracy of identification difference when mark object has incompleteness, is at least partially obscured；Cuda programming can be used, realize GPU, CPU parallel computation
To accelerate deep learning calculating speed, non-mark image data is largely acquired for training neural network, automatically using mixing
Encoder obtains accurate, strong antijamming capability image data feature, and monitor layer is added after mixing autocoder, is exported for it
Characteristics of image unify automatic marking, and the characteristic image of mark is input to Faster RCNN neural metwork training data mould
Type is finally realized and is identified and positioned to target by another group of data back training neural network parameter；It is anti-to improve model
Interference performance and its robustness reduce data set training difficulty, improve model training speed and quality, make neural network convenient for needle
It is easy to use to different targets to be identified, there is important directive significance to image procossing and target identification technology field.
Detailed description of the invention:
Fig. 1 is target identification method flow diagram in embodiment；
Fig. 2 is Target Identification Unit schematic diagram in embodiment；
Fig. 3 is acquisition module schematic diagram in embodiment；
Fig. 4 is coding module schematic diagram in embodiment；
Fig. 5 is supervised classification submodule schematic diagram in embodiment；
Fig. 6 is target identification schematic illustration in embodiment；
Fig. 7 is that autocoder schematic illustration is mixed in embodiment；
Fig. 8 is label annotation flow journey figure in embodiment；
Fig. 9 is Faster RCNN neural network model schematic diagram in embodiment.
Specific embodiment:
To make the object, technical solutions and advantages of the present invention clearer, understand, with reference to the accompanying drawing with technical solution pair
The present invention is described in further detail.
At present in image processing techniques, data set training is carried out by supervised learning, Primary Stage Data collection mark, production can be spent
Take a large amount of manpower and material resources；Unsupervised learning is used alone and be easy to cause the problems such as being not easy to applied to target detection, over-fitting.
For this purpose, the embodiment of the present invention, shown in Figure 1, a kind of object detection method for parallel manipulator human visual system is provided, is wrapped
Containing following content:
S101, acquisition target image, and collected target image is divided into training dataset and test data set；
S102, mixing autocoder is built, concentrates image to be sent into mixing autocoder in batches training data and carries out
Denoising and sparse processing obtain the image data for having tag along sort；
S103, Faster RCNN neural network progress model training will be sent into tag along sort image data；
S104, two groups of target images to be detected for concentrating test data are sent into trained Faster RCNN nerve
Model measurement is carried out in network；
S105, the Faster RCNN neural network model after test is embedded in parallel manipulator human visual system, passes through work
Industry video camera carries out object recognition and detection to image to be detected collected in field range.
Preferably, it acquires in target image, which is divided into two groups, wherein one group is every figure
Piece only includes an object to be identified, and another group is there are a variety of difference objects to be identified in every picture；And it is to be checked by every group
It surveys target image and is divided into training data and test data, every group of corresponding image data is inputted into training dataset respectively
And test data set.Picture in training set comprising target is obtained into an accurately mesh by Background difference with pure background picture
Candidate frame is marked, image is sheared according to candidate frame, and the target that folder name is referred to as tag along sort and shearing obtains
In conjunction with acquisition has the data set of label.Each group of classification obtains the data set of label all in accordance with this mode.Denoising and sparse place
Reason has raw data set feature more abstract, effective, sparse, to improve model training speed and model generalization ability.Band
Have tag along sort image data to be sent into CNN neural network progress model training is made with quick obtaining fully-connected network weight
Faster RCNN weight training is more optimized, accelerates Faster RCNN model training process and precision.
Picture in training set comprising target is obtained into an accurately target time by Background difference with pure background picture
Frame is selected, image is sheared according to candidate frame, and folder name is referred to as tag along sort in conjunction with the target that shearing obtains,
Obtain the data set for having label.Each group of classification obtains the data set of label all in accordance with this mode., by mixing autocoding
Device processing carries out the sparse processing of noise reduction to Target Photo, data set is made to show more abstract, effective, sparse feature.Each group of class
The data set of label is not obtained all in accordance with this mode.Mixing autocoder is built, training data is concentrated to the number for having label
According to collection mixing autocoder is sent into according to labeled packet in batches and carries out denoising and sparse processing, there is raw data set feature
It is more abstract, effective, sparse, to improve model training speed and model generalization ability.It will be sent with tag along sort image data
Enter CNN neural network progress model training keeps Faster RCNN weight training more excellent with quick obtaining fully-connected network weight
Change, accelerates Faster RCNN model training process and precision.Will with tag along sort image data be sent into CNN neural network into
Row model training keeps Faster RCNN weight training more optimized with quick obtaining fully-connected network weight, accelerates Faster
RCNN model training process and precision.Initial model data are sent into Faster RCNN neural network and carry out model training, are obtained
Final target detection model.Two groups of target images to be detected that test data is concentrated are dilute by the noise reduction of mixing autocoder
Processing is dredged, output is sent into trained Faster RCNN neural network and carries out model measurement.Label for labelling flow chart is such as
Shown in Fig. 8, Faster RCNN neural network model is as shown in figure 9, Faster RCNN neural network training process: (1) training
CNN neural network is quickly obtained network weight；(2) training RPN network and shared convolutional layer, and every picture is predicted,
2000 candidate regions；(3) the CNN network weight generated using the first step and second obtained candidate region training Fast
RCNN and shared convolutional layer；(4) fixed convolutional layer, training RPN simultaneously predict 2000 candidate regions；(5) fixed convolutional layer, instruction
Practice Fast RCNN.
Above-mentioned, the mixing autocoder includes first layer denoising self-encoding encoder, second layer denoising self-encoding encoder, the
Three layers of sparse self-encoding encoder and the monitor layer that every layer of output end is set；The output characteristics of image of three layer coders is respectively as right
The input for answering output end monitor layer by image difference, filters out small area pixel etc. and obtains target area, and every group according to file
Name acquiring label obtains the image data for having tag along sort.
Above-mentioned, the monitor layer classifies automatically to output characteristics of image and adds tag along sort, obtains to have and divide
The image data of class label.For the image data of input, Faster RCNN neural network is by preceding to, back-propagation algorithm
Carry out model training test.
Based on above-mentioned target identification method, the present invention also provides a kind of target inspections for parallel manipulator human visual system
Device is surveyed, it is shown in Figure 2, include: acquisition module 101, coding module 102, training module 103, test module 104 and target
Identification module 105, wherein
Acquisition module 101 is divided into training dataset and test for acquiring target image, and by collected target image
Data set；
Training data is concentrated image to be sent into mixing in batches automatic by coding module 102 for building mixing autocoder
Encoder carries out denoising and sparse processing, obtains the image feature data for having tag along sort；
Training module 103 is sent into CNN and Faster with tag along sort image feature data for export monitor layer
RCNN neural network carries out model training；
Test module 104, two groups of target images to be detected for concentrating test data are sent into trained
Model measurement is carried out in Faster RCNN neural network；
Target identification module 105, for the Faster RCNN neural network model insertion parallel robot view after testing
In feel system, object recognition and detection is carried out to image to be detected collected in field range by industrial camera.
Shown in Figure 3 in above-mentioned device, acquisition module 101 includes: acquisition submodule 1001 and grouping submodule
1002, wherein
Submodule 1001 is acquired, collects target data for shooting, which includes that every picture only includes one
There are the multi-Target Image data of a variety of objects Bu Tong to be identified in every picture for the single goal image data of object to be identified；
It is grouped submodule 1002, for being divided into the target data being collected into containing physical quantities to be identified according in picture
Two groups, wherein one group is single goal image data set, and another group is multi-Target Image data group；And by every group of target to be detected
Image is divided into training data and test data, and every group of corresponding image data is inputted training dataset and test respectively
Data set.
It is shown in Figure 4 in above-mentioned device, the coding module 102 include the first denoising encoding submodule 2001,
Second denoising encoding submodule 2002, sparse coding submodule 2003 and supervised classification submodule 2004, wherein
First denoising encoding submodule 2001, for being that image data adds man-made noise using denoising autocoder,
Image training characteristics are constructed, more improve robustness for following model training；
Second denoising encoding submodule 2002, for being that image data adds man-made noise using denoising autocoder,
Image training characteristics are constructed, more improve robustness for following model training；
Sparse coding submodule 2003 obtains more sparse for carrying out image real time transfer using sparse self-encoding encoder
Output image data, improve the abstractness of data, reduce model training data volume, thus accelerate target detection model training and
Recognition speed；
Supervised classification submodule 2004, for by adding in denoising both autocoder and sparse self-encoding encoder output end
Add monitor layer, obtains the image feature data eventually with tag along sort.
Shown in Figure 5 in above-mentioned device, the supervised classification submodule 2004 includes 401 He of data categorization module
Add label model 402, wherein
Data categorization module 401, for carrying out data classification to output characteristics of image；
Label model 402 is added, for adding label automatically to sorted image feature data, is obtained with contingency table
The image feature data of label.
Shown in Figure 6 in the present invention, firstly, acquiring target image to be detected, Image Acquisition is divided into two groups, first group
Only to include an object to be identified in every picture, second group is there are several difference objects to be identified in every picture.Often
Group picture is all divided into training dataset and test data set；Mixing autocoder is built, first layer is denoising self-encoding encoder, the
Two layers are denoising self-encoding encoder, and third layer is sparse self-encoding encoder；Picture classification will not marked, and to be sent into above-mentioned mixing in batches automatic
Encoder obtains output feature.After mixing autocoder the last layer plus monitor layer, monitor layer input are compiled for mixing is automatic
Code device output, which is that autocoder output is unified adds tag along sort automatically；There is label picture defeated above-mentioned output
Enter to Faster RCNN neural network, training data model；Model is tested with two groups of pictures, first group is single goal
Image, second group is multi-Target Image；It is sent into image to be detected for the network model of generation, it is identified and positioned.It gives
Target coordinate position out, guidance parallel robot move to target position.Acquire target image in, to target to be detected into
Row single goal shoots picture, different targets is placed under the file of respective classes, file name is identical as target designation.
Multi-target detection is carried out to target to be detected, and is put in individual files folder.Above-mentioned file is to be used as data set.It takes
Build autocoder, and do following improvement to autocoder: shown in Figure 7, changing single layer autocoder is that depth is self-editing
Code device, the autocoder number of plies are set as three layers, and each layer uses different classes of autocoder, and first two layers automatic using denoising
Encoder adds man-made noise for image and passes through the image of denoising autocoder construction strong antijamming capability, more robustness
Training characteristics, third layer use sparse self-encoding encoder, reduce data parameters, accelerate following model training difficulty, reduce over-fitting.
Gained unlabeled data collection is classified and is sent into the mixing autocoder that step 2 is built, it is more characteristic to extract image
The processed characteristic image for mixing autocoder output is input to the supervision for mixing autocoder addition by feature
Layer carries out automatic classification annotation for these feature images, to obtain the apparent data set of label, feature.Improve data mould
The generalization ability of type.
The present invention solves traditional detection method data set and makes heavy workload, computationally intensive, or mesh close for color
The problem of accuracy of identification difference when mark object has incompleteness, is at least partially obscured；Cuda programming can be used, realize GPU, CPU parallel computation
To accelerate deep learning calculating speed.Largely acquire non-mark image data for training neural network, automatically using mixing
Encoder obtains accurate, strong antijamming capability image data feature, and monitor layer is added after mixing autocoder, is exported for it
Characteristics of image unify automatic marking, and the characteristic image of mark is input to Faster RCNN neural metwork training data mould
Type is finally realized and is identified and positioned to target by another group of data back training neural network parameter.Early period uses nothing
Supervision mixing autocoder handles image data, can obtain strong antijamming capability, characteristics of image accurate, data volume is small,
Faster RCNN classifier is inputted again, recognition training is carried out to it, to improve model anti-interference ability and its robustness, is dropped
Low data set training difficulty, improves model training speed and quality, and neural network is made to be convenient for being easy to for different targets to be identified
It reuses.
Unless specifically stated otherwise, the opposite step of the component and step that otherwise illustrate in these embodiments, digital table
It is not limit the scope of the invention up to formula and numerical value.
Based on above-mentioned method, the embodiment of the present invention also provides a kind of server, comprising: one or more processors；It deposits
Storage device, for storing one or more programs, when one or more of programs are executed by one or more of processors,
So that one or more of processors realize above-mentioned method.
Based on above-mentioned method, the embodiment of the present invention also provides a kind of computer-readable medium, is stored thereon with computer
Program, wherein the program realizes above-mentioned method when being executed by processor.
The technical effect and preceding method embodiment phase of device provided by the embodiment of the present invention, realization principle and generation
Together, to briefly describe, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
It with the specific work process of device, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
In all examples being illustrated and described herein, any occurrence should be construed as merely illustratively, without
It is as limitation, therefore, other examples of exemplary embodiment can have different values.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
The flow chart and block diagram in the drawings show the system of multiple embodiments according to the present invention, method and computer journeys
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, section or code of table, a part of the module, section or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two continuous boxes can actually base
Originally it is performed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that
It is the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, can uses and execute rule
The dedicated hardware based system of fixed function or movement is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes by another way.The apparatus embodiments described above are merely exemplary, for example, the division of the unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, in another example, multiple units or components can
To combine or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or beg for
The mutual coupling, direct-coupling or communication connection of opinion can be through some communication interfaces, device or unit it is indirect
Coupling or communication connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in the executable non-volatile computer-readable storage medium of a processor.Based on this understanding, of the invention
Technical solution substantially the part of the part that contributes to existing technology or the technical solution can be with software in other words
The form of product embodies, which is stored in a storage medium, including some instructions use so that
One computer equipment (can be personal computer, server or the network equipment etc.) executes each embodiment institute of the present invention
State all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can be with
Store the medium of program code.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (11)
1. a kind of object detection method for parallel manipulator human visual system, which is characterized in that include following content:
Target image is acquired, and collected target image is divided into training dataset and test data set；
Mixing autocoder is built, has the data set of label to be sent into mixing in batches certainly according to labeled packet training data concentration
Dynamic encoder carries out denoising and sparse processing；
CNN neural network will be sent into tag along sort image data carry out model training；
Initial model data are sent into Faster RCNN neural network and carry out model training, obtain final target detection model；
Two groups of target images to be detected that test data is concentrated will be exported by the sparse processing of noise reduction of mixing autocoder
It is sent into trained Faster RCNN neural network and carries out model measurement；
By in the Faster RCNN neural network model insertion parallel manipulator human visual system after test, pass through industrial camera
Object recognition and detection is carried out to image to be detected collected in field range.
2. the object detection method according to claim 1 for parallel manipulator human visual system, which is characterized in that acquisition
In target image, which is divided into two groups, wherein one group only includes an object to be identified for every picture
Body, another group is there are a variety of difference objects to be identified in every picture；And every group of target image to be detected is divided into
Every group of corresponding image data is inputted training dataset and test data set by training data and test data respectively.
3. the object detection method according to claim 1 for parallel manipulator human visual system, which is characterized in that described
Mix autocoder include first layer denoising self-encoding encoder, the second layer denoising self-encoding encoder, the sparse self-encoding encoder of third layer and
Three layers of monitor layer of every layer of output end are set；The output characteristics of image of encoder is defeated respectively as corresponding output end monitor layer
Enter, obtain target area, carries out tag along sort processing.
4. the object detection method according to claim 3 for parallel manipulator human visual system, which is characterized in that described
Monitor layer classifies automatically to output characteristics of image and adds tag along sort, and the image data for having tag along sort is obtained.
5. the object detection method according to claim 1 for parallel manipulator human visual system, which is characterized in that be directed to
The image data of input, Faster RCNN neural network carry out model training test to, back-propagation algorithm by preceding.
6. a kind of object detecting device for parallel manipulator human visual system is, characterized by comprising: acquisition module, coding
Module, training module, test module and target identification module, wherein
Acquisition module is divided into training dataset and test data set for acquiring target image, and by collected target image；
Training data is concentrated image to be sent into mixing autocoder in batches by coding module for building mixing autocoder
Denoising and sparse processing are carried out, it is final to obtain the image feature data for having tag along sort；
Training module is sent into CNN, Faster RCNN mind with tag along sort image feature data for export monitor layer
Model training is carried out through network；
Test module, two groups of target images to be detected for concentrating test data are sent into trained Faster RCNN
Model measurement is carried out in neural network；
Target identification module is embedded in parallel manipulator human visual system for the Faster RCNN neural network model after testing
In, object recognition and detection is carried out to image to be detected collected in field range by industrial camera.
7. the object detecting device according to claim 6 for parallel manipulator human visual system, which is characterized in that acquisition
Module includes: acquisition submodule and grouping submodule, wherein
Submodule is acquired, collects target data for shooting, which includes that every picture only includes an object to be identified
There are the multi-Target Image data of a variety of objects Bu Tong to be identified in every picture for the single goal image data of body；
It is grouped submodule, for the target data being collected into be divided into two groups containing physical quantities to be identified according in picture,
In, one group is single goal image data set, and another group is multi-Target Image data group；And every group of target image to be detected is distinguished
It is divided into training data and test data, every group of corresponding image data is inputted into training dataset and test data set respectively.
8. the object detecting device according to claim 6 for parallel manipulator human visual system, which is characterized in that described
Coding module include the first denoising encoding submodule, the second denoising encoding submodule, sparse coding submodule and supervised classification
Submodule, wherein
First denoising encoding submodule, for being that image data adds man-made noise using denoising autocoder；
Second denoising encoding submodule, for being that image data adds man-made noise using denoising autocoder；
Sparse coding submodule, for carrying out image real time transfer using sparse self-encoding encoder；
Supervised classification submodule, for adding monitor layer by the output end in denoising autocoder and sparse self-encoding encoder,
Obtain the image feature data eventually with tag along sort.
9. the object detecting device according to claim 6 for parallel manipulator human visual system, which is characterized in that described
Supervised classification submodule includes data categorization module and addition label model, wherein
Data categorization module, for carrying out data classification to output characteristics of image；
Label model is added, for adding label automatically to sorted image feature data, obtains the figure with tag along sort
As characteristic.
10. a kind of server, comprising: one or more processors；Storage device works as institute for storing one or more programs
It states one or more programs to be executed by one or more of processors, so that one or more of processors realize that right is wanted
Method described in asking 1.
11. a kind of computer-readable medium, is stored thereon with computer program, wherein the realization when program is executed by processor
Method described in claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811509164.1A CN109785333A (en) | 2018-12-11 | 2018-12-11 | Object detection method and device for parallel manipulator human visual system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811509164.1A CN109785333A (en) | 2018-12-11 | 2018-12-11 | Object detection method and device for parallel manipulator human visual system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109785333A true CN109785333A (en) | 2019-05-21 |
Family
ID=66495820
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811509164.1A Pending CN109785333A (en) | 2018-12-11 | 2018-12-11 | Object detection method and device for parallel manipulator human visual system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785333A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079581A (en) * | 2019-12-03 | 2020-04-28 | 广州久邦世纪科技有限公司 | Method and device for identifying human skin |
| CN111126359A (en) * | 2019-11-15 | 2020-05-08 | 西安电子科技大学 | High-definition image small target detection method based on self-encoder and YOLO algorithm |
| CN111199523A (en) * | 2019-12-24 | 2020-05-26 | 深圳供电局有限公司 | Power equipment identification method and device, computer equipment and storage medium |
| CN111460909A (en) * | 2020-03-09 | 2020-07-28 | 兰剑智能科技股份有限公司 | Vision-based goods location management method and device |
| CN113378944A (en) * | 2021-06-17 | 2021-09-10 | 北京博创联动科技有限公司 | Agricultural machinery operation mode recognition model training method and device and terminal equipment |
| CN113688825A (en) * | 2021-05-17 | 2021-11-23 | 海南师范大学 | AI intelligent garbage recognition and classification system and method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203506A (en) * | 2016-07-11 | 2016-12-07 | 上海凌科智能科技有限公司 | A kind of pedestrian detection method based on degree of depth learning art |
| CN107610086A (en) * | 2016-07-12 | 2018-01-19 | 天津超众机器人科技有限公司 | Industrial parallel robot rapid visual detection algorithm based on bionic compound eye structure |
| CN108010060A (en) * | 2017-12-06 | 2018-05-08 | 北京小米移动软件有限公司 | Object detection method and device |
| CN108229665A (en) * | 2018-02-02 | 2018-06-29 | 上海建桥学院 | A kind of the System of Sorting Components based on the convolutional neural networks by depth |
-
2018
- 2018-12-11 CN CN201811509164.1A patent/CN109785333A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203506A (en) * | 2016-07-11 | 2016-12-07 | 上海凌科智能科技有限公司 | A kind of pedestrian detection method based on degree of depth learning art |
| CN107610086A (en) * | 2016-07-12 | 2018-01-19 | 天津超众机器人科技有限公司 | Industrial parallel robot rapid visual detection algorithm based on bionic compound eye structure |
| CN108010060A (en) * | 2017-12-06 | 2018-05-08 | 北京小米移动软件有限公司 | Object detection method and device |
| CN108229665A (en) * | 2018-02-02 | 2018-06-29 | 上海建桥学院 | A kind of the System of Sorting Components based on the convolutional neural networks by depth |
Non-Patent Citations (2)
| Title |
|---|
| 张成刚等: "一种稀疏降噪自编码神经网络研究", 《内蒙古民族大学学报（自然科学版）》 * |
| 贾文娟等: "自编码器理论与方法综述", 《计算机系统应用》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111126359A (en) * | 2019-11-15 | 2020-05-08 | 西安电子科技大学 | High-definition image small target detection method based on self-encoder and YOLO algorithm |
| CN111079581A (en) * | 2019-12-03 | 2020-04-28 | 广州久邦世纪科技有限公司 | Method and device for identifying human skin |
| CN111199523A (en) * | 2019-12-24 | 2020-05-26 | 深圳供电局有限公司 | Power equipment identification method and device, computer equipment and storage medium |
| CN111199523B (en) * | 2019-12-24 | 2023-08-25 | 深圳供电局有限公司 | Power equipment identification method, device, computer equipment and storage medium |
| CN111460909A (en) * | 2020-03-09 | 2020-07-28 | 兰剑智能科技股份有限公司 | Vision-based goods location management method and device |
| CN113688825A (en) * | 2021-05-17 | 2021-11-23 | 海南师范大学 | AI intelligent garbage recognition and classification system and method |
| CN113378944A (en) * | 2021-06-17 | 2021-09-10 | 北京博创联动科技有限公司 | Agricultural machinery operation mode recognition model training method and device and terminal equipment |
| CN113378944B (en) * | 2021-06-17 | 2022-02-18 | 北京博创联动科技有限公司 | Agricultural machinery operation mode recognition model training method and device and terminal equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785333A (en) | Object detection method and device for parallel manipulator human visual system | |
| Huang et al. | Point cloud labeling using 3d convolutional neural network | |
| WO2020046213A1 (en) | A method and apparatus for training a neural network to identify cracks | |
| CN113361645B (en) | Target detection model construction method and system based on meta learning and knowledge memory | |
| CN114821014B (en) | Multi-mode and countermeasure learning-based multi-task target detection and identification method and device | |
| Li et al. | Visual stability prediction for robotic manipulation | |
| Sapijaszko et al. | An overview of recent convolutional neural network algorithms for image recognition | |
| CN112734803B (en) | Single target tracking method, device, equipment and storage medium based on character description | |
| Yan et al. | Underwater object detection algorithm based on attention mechanism and cross-stage partial fast spatial pyramidal pooling | |
| CN115810157A (en) | Unmanned aerial vehicle target detection method based on lightweight feature fusion | |
| Vinicki et al. | Using convolutional neural networks for determining reticulocyte percentage in cats | |
| Isa et al. | CNN transfer learning of shrimp detection for underwater vision system | |
| Mirani et al. | Object recognition in different lighting conditions at various angles by deep learning method | |
| Kaya et al. | IsVoNet8: a proposed deep learning model for classification of some fish species | |
| Wulandari et al. | A comparison of deep learning approach for underwater object detection | |
| Dong et al. | Visual Detection Algorithm for Enhanced Environmental Perception of Unmanned Surface Vehicles in Complex Marine Environments | |
| CN115620122A (en) | Training method of neural network model, image re-recognition method and related equipment | |
| Chan et al. | Real-time Detection of Aquarium Fish Species Using YOLOv4-tiny on Raspberry Pi 4 | |
| CN115601551A (en) | Object identification method and device, storage medium and electronic equipment | |
| Zhang et al. | Research on ship target detection based on improved YOLOv5 algorithm | |
| CN111914949A (en) | Zero sample learning model training method and device based on reinforcement learning | |
| Shishkin et al. | Implementation of yolov5 for detection and classification of microplastics and microorganisms in marine environment | |
| Andriyanov et al. | Optimization of the computer vision system for the detection of moving objects | |
| Visalatchi et al. | Intelligent Vision with TensorFlow using Neural Network Algorithms | |
| Kushwaha et al. | Analysis of CNN Model with Traditional Approach and Cloud AI based Approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |
|
| RJ01 | Rejection of invention patent application after publication |