CN109063761A - Diffuser dropping detection method, device and electronic equipment - Google Patents
Diffuser dropping detection method, device and electronic equipment Download PDFInfo
- Publication number
- CN109063761A CN109063761A CN201810810454.3A CN201810810454A CN109063761A CN 109063761 A CN109063761 A CN 109063761A CN 201810810454 A CN201810810454 A CN 201810810454A CN 109063761 A CN109063761 A CN 109063761A
- Authority
- CN
- China
- Prior art keywords
- feature
- sample
- new samples
- feature extraction
- diffuser
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of diffuser dropping detection method, device and electronic equipments, are related to 3D field of camera technology, this method comprises: obtaining image to be detected by photographic device；By image to be detected input feature extraction network trained in advance, so that feature extraction network extracts the feature of image to be detected；Wherein, feature extraction network is generated by new samples collection training；New samples collection is constituted by giving the new samples group that sample is combined；By feature input classifier trained in advance, classification results are obtained；Determine whether that diffuser, which occurs, to fall off according to classification results.The embodiment of the present invention can carry out more efficient diffuser to photographic device and fall off detection, and accuracy in detection is higher.
Description
Technical field
The present invention relates to 3D field of camera technology, more particularly, to a kind of diffuser dropping detection method, device and electronics
Equipment.
Background technique
With the extensive use of 3D mould group, more and more mobile terminals can have 3D camera module, such as common fly
Row time (Time-of-flight) mould group technology issues modulated near infrared light by sensor, meets object back reflection, passes
Sensor is by calculating light transmitting and reflection interval difference or phase difference, come the distance of scenery of being taken that converts, to generate depth letter
Furthermore breath shoots in conjunction with traditional camera, the three-D profile of object can be represented in different colors to the landform of different distance
Figure mode shows.
Above-mentioned camera module is usually mounted with diffuser (Diffuser, also referred to as diffusion sheet or scattering before infrared light supply
Device), enable infrared light to be uniformly irradiated to entire photographed scene.If infrared light supply issues infrared without diffuser
Light can be gathered into a branch of, cause 3D camera that can not normally perceive the depth of subject in scene.In actual use,
3D camera module is collided or shakes and has certain probability and the problem of diffuser falls off occur, so that 3D camera module is lost
Effect.
For in the prior art, the poor problem of the precision that detection diffuser falls off not yet proposes effective solution at present
Scheme.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of diffuser dropping detection method, device and electronic equipment, it can
To improve the precision that detection diffuser falls off.
To achieve the goals above, technical solution used in the embodiment of the present invention is as follows:
In a first aspect, the diffuser, which is set to, to be taken the photograph the embodiment of the invention provides a kind of diffuser dropping detection method
As device, this method comprises: obtaining image to be detected by the photographic device；Described image to be detected is inputted into training in advance
Feature extraction network so that the feature extraction network extracts the feature of described image to be detected；Wherein, the feature extraction
Network is generated by new samples collection training；The new samples collection is constituted by giving the new samples group that sample is combined
's；By feature input classifier trained in advance, classification results are obtained；Determine whether to expand according to the classification results
Scattered device falls off.
Further, the method also includes: the new samples group of one that the new samples are concentrated inputs the feature
Extract network；The new samples group includes at least two given samples；It is extracted respectively often by the feature extraction network
The feature of a given sample；If two given samples belong to the sample of the same category, random sample will be given described in two
This feature carries out minimum processing, with the training feature extraction network；The classification includes the non-classification and de- of falling off
Fall classification；If two given samples are not belonging to the sample of the same category, by the feature of two given samples
Maximization processing is carried out, with the training feature extraction network；The new samples group that the new samples are concentrated is sequentially input, until institute
Stopping when stating feature extraction network convergence.
Further, the step of new samples group input feature vector that the new samples are concentrated extracts network,
It include: that the sample that falls off, a normal sample and an attack sample are formed the new samples group input feature to mention
Take network；The feature by two given samples carries out the step of minimizing processing, comprising: by the attack sample
The feature of this and the normal sample carries out minimum processing；The feature by two given samples carries out
The step of maximizing processing, comprising: the feature of fall off sample and the normal sample is subjected to maximization processing；It will
It is described fall off sample and it is described attack sample the feature carry out maximization processing.
Further, the step of new samples group input feature vector that the new samples are concentrated extracts network,
It include: that the given sample of random selection two forms the new samples group, and the new samples group is inputted the feature and is mentioned
Take network.
Further, the feature by two training samples carries out the step of minimizing processing, comprising: calculates
Distance between the vector of two features, by the distance minimization, to carry out minimum processing.
Further, the feature by two training samples carries out the step of maximizing processing, comprising: calculates
Distance between the vector of two features, and the distance is maximized, to carry out maximization processing.
Further, the method also includes: by the feature extraction network trained in advance, to the sample that falls off, normal
Sample and attack sample carry out feature extraction；The feature of the sample that falls off is divided into the classification that falls off, by the normal sample
It is divided into the non-classification that falls off with the feature of attack sample, the classifier is trained.
Further, the method also includes: when determine occur diffuser fall off when, carry out warning reminding.
Second aspect, falls off detection device the embodiment of the invention also provides a kind of diffuser, and the diffuser is set to
Photographic device, the device include: acquisition module, for obtaining image to be detected by the photographic device；Characteristic extracting module,
For the feature extraction network that the input of described image to be detected is trained in advance so that the feature extraction network extract it is described to
The feature of detection image；Wherein, the feature extraction network is generated by new samples collection training；The new samples collection be by
What the new samples group that random sample is originally combined was constituted；Categorization module, for the classification that feature input is trained in advance
Device obtains classification results；Judgment module, for determining whether that diffuser, which occurs, to fall off according to the classification results.
The third aspect, the embodiment of the invention provides a kind of electronic equipment, including memory and processor, the memories
In be stored with the computer program that can be run on the processor, the processor realizes the when executing the computer program
On the one hand the step of described in any item methods.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage
Computer program is stored on medium, the computer program is executed when being run by processor described in above-mentioned any one of first aspect
Method the step of.
The embodiment of the invention provides a kind of diffuser dropping detection method, device and electronic equipments, can be by preparatory
Trained feature extraction network handles detection image carries out feature extraction, which obtained by photographic device to be detected
, this feature, which extracts network, to be trained by giving the new samples collection that sample is combined, and is obtained in extraction
Classify after feature, and determine whether that there is a situation where diffusers to fall off, more efficient diffusion can be carried out to photographic device
Device falls off detection, and accuracy in detection is higher.
Other feature and advantage of the disclosure will illustrate in the following description, alternatively, Partial Feature and advantage can be with
Deduce from specification or unambiguously determine, or by implement the disclosure above-mentioned technology it can be learnt that.
To enable the above objects, features, and advantages of the disclosure to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of structural schematic diagram of processing equipment provided in an embodiment of the present invention；
Fig. 2 is a kind of flow chart of diffuser dropping detection method provided in an embodiment of the present invention；
Fig. 3 is a kind of flow chart of feature extraction network training method provided in an embodiment of the present invention；
Fig. 4 is a kind of schematic diagram of given sample provided in an embodiment of the present invention；
Fig. 5 is the schematic diagram that a kind of training characteristics provided in an embodiment of the present invention extract network；
Fig. 6 is the schematic diagram that another training characteristics provided in an embodiment of the present invention extract network；
Fig. 7 is that a kind of diffuser provided in an embodiment of the present invention falls off the structural block diagram of detection device.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Technical solution be clearly and completely described, it is clear that described embodiments are some of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
With application of the 3D detection technology on intelligent terminal, falls off to detect to become to the diffuser of 3D camera module and newly ask
Topic.Intelligent terminal manufacturer or camera module manufacturer require especially to use up 99.9% or more for the detection accuracy of diffuser
The mistake that amount reduces non-defective unit is known.When intelligent terminal user actually uses intelligent terminal, whether fall off there is also detection diffuser
It needs.The method precision whether existing detection diffuser falls off is poor, is unable to satisfy manufacturer or user demand.It is asked to improve this
Topic, the embodiment of the invention provides a kind of diffuser dropping detection method, device and electronic equipments, below to the embodiment of the present invention
It describes in detail.
Embodiment one:
Firstly, describing diffuser dropping detection method, device and electronics for realizing the embodiment of the present invention referring to Fig.1
The exemplary electronic device 100 of equipment.
The structural schematic diagram of a kind of electronic equipment as shown in Figure 1, electronic equipment 100 include one or more processors
102, one or more storage devices 104, input unit 106, output device 108 and image collecting device 110, these components
It is interconnected by bindiny mechanism's (not shown) of bus system 112 and/or other forms.It should be noted that electronic equipment shown in FIG. 1
100 component and structure be it is illustrative, and not restrictive, as needed, the electronic equipment also can have other
Component and structure.
The processor 102 can use digital signal processor (DSP), field programmable gate array (FPGA), can compile
At least one of journey logic array (PLA) example, in hardware realizes that the processor 102 can be central processing unit
(CPU) or one or more of the processing unit of other forms with data-handling capacity and/or instruction execution capability
Combination, and can control other components in the electronic equipment 100 to execute desired function.
The storage device 104 may include one or more computer program products, and the computer program product can
To include various forms of computer readable storage mediums, such as volatile memory and/or nonvolatile memory.It is described easy
The property lost memory for example may include random access memory (RAM) and/or cache memory (cache) etc..It is described non-
Volatile memory for example may include read-only memory (ROM), hard disk, flash memory etc..In the computer readable storage medium
On can store one or more computer program instructions, processor 102 can run described program instruction, to realize hereafter institute
The client functionality (realized by processor) in the embodiment of the present invention stated and/or other desired functions.In the meter
Can also store various application programs and various data in calculation machine readable storage medium storing program for executing, for example, the application program use and/or
The various data etc. generated.
The input unit 106 can be the device that user is used to input instruction, and may include keyboard, mouse, wheat
One or more of gram wind and touch screen etc..
The output device 108 can export various information (for example, image or sound) to external (for example, user), and
It and may include one or more of display, loudspeaker etc..
Described image acquisition device 110 can shoot the desired image of user (such as photo, video etc.), and will be clapped
The image taken the photograph is stored in the storage device 104 for the use of other components.Described image acquisition device 110 is imaged including 3D
Mould group, which includes diffuser.
Illustratively, show for realizing diffuser dropping detection method according to an embodiment of the present invention, apparatus and system
Example electronic equipment may be implemented as the intelligent terminals such as smart phone, tablet computer, computer.
Embodiment two:
A kind of flow chart of diffuser dropping detection method shown in Figure 2, the diffuser are set to photographic device, should
The electronic equipment that method can be provided by previous embodiment executes, and this method specifically comprises the following steps:
Step S202 obtains image to be detected by photographic device.
The photographic device can be both installed on intelligent terminal, be also possible to independently to use.It is carried out to photographic device
Diffuser falls off when detecting, and image is acquired by the photographic device, as image to be detected.For example, being connect by above-mentioned intelligent terminal
It receives and stores the image to be detected or other external equipments connecting with the photographic device receive and store the image to be detected.
Step S204, by image to be detected input feature extraction network trained in advance, so that feature extraction network extracts
The feature of image to be detected.
Wherein, feature extraction network is generated by giving the new samples collection training that sample is combined.By feature
The mode that network carries out deep learning is extracted, the accuracy of feature extraction is higher, and does not need to manually adjust parameter, but operation
Speed is relatively slow, needs in advance to be trained model.
The case where being fallen off due to diffuser be not common, and there is a problem of falling off, sample is few, needs more than disposition.
Sample given herein above includes falling off sample and the non-sample that falls off, and the present embodiment can will fall off sample and the non-sample progress that falls off
Combination, for example, the sample that falls off is combined with a non-sample that falls off, the sample that falls off is combined with two non-samples that fall off, two
A fall off sample combination or two non-modes such as sample combination that fall off, obtain new samples group, the quantity of the new samples group is greater than
The quantity for sample and the normal sample of falling off and, the new samples collection that above-mentioned new samples group is constituted is as the training of feature extraction network
Sample set.The training samples number of feature extraction network is increased, through the above way so as to only pass through original sample
Notebook data, which obtains, more accurately extracts model.
Feature input classifier trained in advance is obtained classification results by step S206.
In training classifier, the sample of input is the feature that features described above extracts that network extraction training sample obtains, can
Each training sample is expressed as a feature vector.The feature vector for corresponding to the sample that falls off is labeled as one kind, it will be right
Ying Yufei fall off sample feature vector labeled as another kind of, classifier is trained by the two.Wherein, the sample that falls off is
The image that photographic device acquires when diffuser falls off；The non-sample that falls off includes normal sample and attack sample, which is
The image of photographic device acquisition when diffuser works normally；The attack sample is that approximation falls off signal, is normally pacified in diffuser
Work is filled, but user's operation is improper or other factors lead to acquired image in the case where shooting exception.
After the completion of above-mentioned classifier training, the feature that can be extracted to feature extraction network is classified.
Step S208 determines whether that diffuser, which occurs, to fall off according to classification results.
After the classification results that classifier exports the feature of image to be detected, camera shooting dress can be determined according to the classification results
Set whether there is a situation where diffusers to fall off.When determining that diffuser, which occurs, to fall off, warning reminding can be carried out, to facilitate intelligence
Can user's needs that go wrong of terminal repair, or facilitate manufacturer that problem photographic device is rejected or repaired.
Above-mentioned diffuser dropping detection method provided in an embodiment of the present invention can pass through feature extraction net trained in advance
Network carries out feature extraction to image to be detected, which obtained by photographic device to be detected, and this feature extracts net
Network is trained by giving the new samples group that sample is combined, and is classified after extraction obtains feature,
And determine whether that there is a situation where diffusers to fall off, more efficient diffuser can be carried out to photographic device and is fallen off detection, and is examined
It is higher to survey accuracy.
The feature extraction network used in the above-mentioned methods is that the feature learning scheme based on small sample is trained,
The flow chart of feature extraction network training method shown in Figure 3, this method specifically comprise the following steps:
Step S302, the new samples group input feature vector that new samples are concentrated extract network.New samples group includes at least
Two given samples.
This feature, which extracts network, can be the existing neural network for carrying out image characteristics extraction.Given sample is carried out
Combination, combination can be random sample combination, such as any two give sample combination or three given sample combinations,
Using multiple given samples after combination as a new samples group, it is input in feature extraction network.By by new samples group
Form can increase the quantity of training sample.
Step S304 extracts the feature of each given sample by feature extraction network respectively.
Before being trained, feature extraction Web vector graphic initial parameter carries out the feature extraction of given sample.
Step S306 carries out the feature of two given samples if two given samples belong to the sample of the same category
Minimum processing extracts network with training characteristics.
Above-mentioned classification includes non-fall off classification and the classification that falls off, by the normal sample and attack sample division in given sample
For the non-classification that falls off, the sample that will fall off is divided into the classification that falls off.The schematic diagram of given sample shown in Figure 4, from left to right
It is successively normal sample, attacks sample and the sample that falls off, by taking IR (Infrared Radiation, infrared ray) image as an example.Its
In, normal sample is the image of photographic device acquisition when diffuser works normally；Attack sample refers to that approximation falls off signal, is to expand
The work of device normal mounting is dissipated, but user's operation is improper or other factors lead to shoot acquired image in the case where exception, one
As have an obvious hot spot in the picture, such as user finger very close to photographic device camera lens；Fall off sample
It originally is the image that photographic device acquires when diffuser falls off, due to, without diffuser, light source being caused to be sent out before the light source of photographic device
Light out be it is relatively narrow a branch of, the image of acquisition also only very small part there are content, other parts without.
Step S308, if two given samples are not belonging to the sample of the same category, by the feature of two given samples into
Row maximization processing extracts network with training characteristics.
To the given sample for belonging to the same category in above-mentioned new samples group, the feature of two given samples is minimized
Processing；To the given sample to belong to a different category in above-mentioned new samples group, the feature of two given samples is carried out at maximization
Reason；To optimize the parameter that features described above extracts network.Maximize and minimum processing, can by the corresponding feature of feature to
The distance of amount carries out, such as: calculate the distance between two feature vectors, maximization will distance maximization, minimum is
By the distance minimization.For example, it is F that sample 1, which obtains feature,1(multi-C vector), it is F that sample 2, which obtains feature,2(the spy of same dimension
Levy vector), maximization allows for distance maximum Max between vector | F1-F2| ^2 is minimised as Min | F1-F2|^2.Pass through maximum
After change/minimum features described above vector distance, so that it may return feature extraction network, the ginseng of final optimization pass feature extraction network
Number, the result that the feature for extracting it is classified are more acurrate.Wherein optimize the method for above-mentioned network, existing network can be used
Optimization method, such as gradient optimization algorithm etc..
Step S310 sequentially inputs the new samples group that the new samples are concentrated, until stopping when feature extraction network convergence.
In the training process, one above-mentioned new samples group of input is trained every time, the feature extraction after being optimized
Network chooses second batch sample again later, the re-optimization on the feature extraction network foundation of above-mentioned optimization, until network convergence is
Only.
In view of given sample may include three kinds: normal sample attacks sample and the sample that falls off, and one can be fallen off
Sample, a normal sample and an attack sample extract network as a new samples group input feature vector and are trained, can also
It is trained using randomly choosing two samples as a new samples group input feature vector extraction network.
Training characteristics shown in Figure 5 extract the schematic diagram of network, with the sample that falls off, a normal sample and one
A attack sample composition new samples group input feature vector is illustrated for extracting network.Feature extraction network carries out above-mentioned sample
Feature extraction respectively obtains the feature that falls off, attack signature and normal characteristics.As shown in figure 5, every two feature is maximized
Or minimum processing, specifically: attack signature and normal characteristics are subjected to minimum processing；The feature that will fall off and normal characteristics into
Row maximization processing；The feature that will fall off and attack signature carry out maximization processing.It is defeated with three samples every time in the above method
Enter, feature extraction network constantly train, until convergence.
Training characteristics shown in Figure 6 extract the schematic diagram of network, are mentioned with randomly selected two sample input feature vectors
It takes and is illustrated for network.Two samples of input every time, in Fig. 6 by taking sample 1 and sample 2 as an example, two samples are from all
It is selected at random in sample, feature extraction network carries out feature extraction to above-mentioned sample, respectively obtains 1 feature of sample and sample 2 is special
Sign.As shown in fig. 6, two features are subjected to maximization or minimum processing, specifically: similar sample carries out minimum processing,
Such as sample 1 and sample 2 belong to above-mentioned non-classification or the classification that falls off of falling off；Inhomogeneity sample is subjected to maximization processing, example
As sample 1 and sample 2 are belonging respectively to above-mentioned non-classification and the classification that falls off of falling off.
It is stated before classifier classified in use, it is also necessary to it is trained, classifier can be existing SVM
(Support Vector Machine, support vector machines), is also possible to other applicable classifiers, and the present embodiment does not make this
It limits.To the training process of classifier, can carry out as follows:
(1) by feature extraction network trained in advance, feature is carried out to the sample that falls off, normal sample and attack sample and is mentioned
It takes；
(2) feature for the sample that falls off is divided into the classification that falls off, the feature of normal sample and attack sample is divided into non-
Fall off classification, is trained to classifier.Wherein, each sample is expressed as a feature vector, by the feature for the sample that falls off to
The feature vector that amount is classified as one kind, normal sample and attack sample is classified as one kind and is trained.
The above method passes through sample and the non-sample that falls off (including normal sample and attack sample) pairs of group of the progress of falling off
It closes, to increase training sample set, and the feature of learn to obtain falling off sample and the non-sample that falls off, the detection that falls off thus is carried out,
Network training can be carried out based on the sample that falls off of small sample, and can efficiently be fallen off detection based on the network.
Embodiment three:
For diffuser dropping detection method provided in embodiment two, the embodiment of the invention provides a kind of diffusers
Fall off detection device, and a kind of diffuser shown in Figure 7 falls off the structural block diagram of detection device, comprising:
Module 701 is obtained, for obtaining image to be detected by photographic device；
Characteristic extracting module 702, for the feature extraction network that image to be detected input is trained in advance, so that feature mentions
Network is taken to extract the feature of image to be detected；Wherein, feature extraction network is generated by new samples collection training；New samples collection is
It is constituted by giving the new samples group that sample is combined；
Categorization module 703 obtains classification results for feature to be mentioned input classifier trained in advance；
Judgment module 704, for determining whether that diffuser, which occurs, to fall off according to classification results.
Above-mentioned diffuser provided in an embodiment of the present invention falls off detection device, can pass through feature extraction net trained in advance
Network carries out feature extraction to image to be detected, which obtained by photographic device to be detected, and this feature extracts net
Network is trained by giving the new samples collection that sample is combined, and is classified after extraction obtains feature,
And determine whether that there is a situation where diffusers to fall off, more efficient diffuser can be carried out to photographic device and is fallen off detection, and is examined
It is higher to survey accuracy.
In one embodiment, above-mentioned apparatus further includes feature extraction network training module, is used for: new samples are concentrated
A new samples group input feature vector extract network；New samples group includes at least two given samples；Pass through feature extraction network
The feature of each given sample is extracted respectively；If two given samples belong to the sample of the same category, by two given samples
Feature carry out minimum processing, with training characteristics extract network；Classification includes normal category and the classification that falls off；If two are given
Random sample is originally not belonging to the sample of the same category, and the feature of two given samples is carried out maximization processing, is extracted with training characteristics
Network；The new samples group that the new samples are concentrated is sequentially input, until stopping when feature extraction network convergence.
Wherein, features described above extracts network training module, is also used to: by the sample that falls off, a normal sample and one
A attack sample composition new samples group input feature vector extracts network；The feature of two given samples is subjected to minimum processing, with
The step of optimizing the parameter of feature extraction network, comprising: the feature for attacking sample and normal sample is subjected to minimum processing；It will
The step of feature of two given samples carries out maximization processing, parameter to optimize feature extraction network, comprising: by the sample that falls off
The feature of this and normal sample carries out maximization processing；The feature of fall off sample and attack sample is subjected to maximization processing.
Features described above extracts network training module, is also used to: the given sample composition new samples group of random selection two, and will
New samples group input feature vector extracts network.Features described above extracts network training module, is also used to: between the vector for calculating two features
Distance, by distance minimization, to carry out minimum processing.Features described above extracts network training module, is also used to: calculating two
Distance between the vector of feature, and will be apart from maximization, to carry out maximization processing.Features described above extracts network training module,
It is also used to: calculating the distance between the vector of two features according to the following formula: | F1-F2 | ^2；Wherein F1 and F2 respectively indicate spy
The feature vector of sign.
In another embodiment, above-mentioned apparatus further includes classifier training module, is used for: passing through spy trained in advance
Sign extracts network, carries out feature extraction to the sample that falls off, normal sample and attack sample；The feature for the sample that falls off is divided into de-
Classification is fallen, the feature of normal sample and attack sample is divided into the non-classification that falls off, classifier is trained.
In another embodiment, above-mentioned apparatus further includes reminding module, for when determine occur diffuser fall off when,
Carry out warning reminding.
The technical effect of device provided by the present embodiment, realization principle and generation is identical with previous embodiment, for letter
It describes, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
In addition, the present embodiment additionally provides a kind of electronic equipment, including memory and processor, being stored in memory can
The computer program run on a processor, processor realize the diffuser that above-described embodiment two provides when executing computer program
The step of dropping detection method.
It is apparent to those skilled in the art that for convenience and simplicity of description, the equipment of foregoing description
Specific work process, can be with reference to the corresponding process in previous embodiment, and details are not described herein.
The present embodiment additionally provides a kind of computer readable storage medium, and meter is stored on the computer readable storage medium
Calculation machine program, the step of method provided by above-described embodiment two is executed when computer program is run by processor.
The computer program of a kind of diffuser dropping detection method, device provided by the embodiment of the present invention and processing equipment
Product, the computer readable storage medium including storing program code, the instruction that said program code includes can be used for executing
Previous methods method as described in the examples, specific implementation can be found in embodiment of the method, and details are not described herein.If the function
It is realized in the form of SFU software functional unit and when sold or used as an independent product, can store computer-readable at one
It takes in storage medium.Based on this understanding, technical solution of the present invention substantially in other words contributes to the prior art
Part or the part of the technical solution can be embodied in the form of software products, which is stored in one
In a storage medium, including some instructions are used so that computer equipment (it can be personal computer, server, or
Network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.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), the various media that can store program code such as magnetic or disk.
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, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of diffuser dropping detection method, which is characterized in that the diffuser is set to photographic device, this method comprises:
Image to be detected is obtained by the photographic device；
By described image to be detected input feature extraction network trained in advance so that the feature extraction network extract it is described to
The feature of detection image；Wherein, the feature extraction network is generated by new samples collection training；The new samples collection be by
What the new samples group that random sample is originally combined was constituted；
By feature input classifier trained in advance, classification results are obtained；
Determine whether that diffuser, which occurs, to fall off according to the classification results.
2. the method according to claim 1, wherein the method also includes:
The new samples group input feature vector that the new samples are concentrated extracts network；The new samples group includes at least two
A given sample；
Extract the feature of each given sample respectively by the feature extraction network；
If two given samples belong to the sample of the same category, the feature of two given samples is carried out most
Smallization processing, with the training feature extraction network；The classification includes non-classification and the classification that falls off of falling off；
If two given samples are not belonging to the sample of the same category, the feature of two given samples is carried out
Maximization processing, with the training feature extraction network；
The new samples group that the new samples are concentrated is sequentially input, until stopping when the feature extraction network convergence.
3. according to the method described in claim 2, it is characterized in that, the new samples that the new samples are concentrated
The step of group input feature vector extracts network, comprising:
One sample that falls off, a normal sample and an attack sample are formed into the new samples group and input the feature extraction
Network；
The feature by two given samples carries out the step of minimizing processing, comprising: by the attack sample
Minimum processing is carried out with the feature of the normal sample；
The feature by two given samples carries out the step of maximizing processing, comprising: by the sample that falls off
Maximization processing is carried out with the feature of the normal sample；By the feature of the fall off sample and the attack sample
Carry out maximization processing.
4. according to the method described in claim 2, it is characterized in that, the new samples that the new samples are concentrated
The step of group input feature vector extracts network, comprising:
The given sample of random selection two forms the new samples group, and the new samples group is inputted the feature extraction
Network.
5. according to the method described in claim 2, it is characterized in that, the feature by two training samples carries out
The step of minimizing processing, comprising:
The distance between the vector of two features is calculated, by the distance minimization, to carry out minimum processing.
6. according to the method described in claim 2, it is characterized in that, the feature by two training samples carries out
The step of maximizing processing, comprising:
The distance between the vector of two features is calculated, and the distance is maximized, to carry out maximization processing.
7. the method according to claim 1, wherein the method also includes:
By the feature extraction network trained in advance, feature extraction is carried out to the sample that falls off, normal sample and attack sample；
The feature of the sample that falls off is divided into the classification that falls off, the feature of the normal sample and attack sample is divided into non-
Fall off classification, is trained to the classifier.
The detection device 8. a kind of diffuser falls off, which is characterized in that the diffuser is set to photographic device, which includes:
Module is obtained, for obtaining image to be detected by the photographic device；
Characteristic extracting module, for the feature extraction network that the input of described image to be detected is trained in advance, so that the feature
Extract the feature that network extracts described image to be detected；Wherein, the feature extraction network is generated by new samples collection training；
The new samples collection is constituted by giving the new samples group that sample is combined；
Categorization module obtains classification results for the classifier that feature input is trained in advance；
Judgment module, for determining whether that diffuser, which occurs, to fall off according to the classification results.
9. a kind of electronic equipment, including memory and processor, it is stored with and can runs on the processor in the memory
Computer program, which is characterized in that the processor realizes any one of claim 1 to 7 when executing the computer program
The step of described method.
10. a kind of computer readable storage medium, computer program, feature are stored on the computer readable storage medium
It is, when the computer program is run by processor the step of 1 to 7 described in any item methods of perform claim requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810810454.3A CN109063761B (en) | 2018-07-20 | 2018-07-20 | Diffuser falling detection method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810810454.3A CN109063761B (en) | 2018-07-20 | 2018-07-20 | Diffuser falling detection method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109063761A true CN109063761A (en) | 2018-12-21 |
| CN109063761B CN109063761B (en) | 2020-11-03 |
Family
ID=64834877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810810454.3A Active CN109063761B (en) | 2018-07-20 | 2018-07-20 | Diffuser falling detection method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109063761B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111310578A (en) * | 2020-01-17 | 2020-06-19 | 上海优加利健康管理有限公司 | Method and device for generating heart beat data sample classification network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160241733A1 (en) * | 2015-02-18 | 2016-08-18 | Xerox Corporation | Led illumination uniformity |
| CN107396094A (en) * | 2017-08-17 | 2017-11-24 | 上海大学 | The automatic testing method of single camera damage towards in multi-cam monitoring system |
| CN108198175A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Detection method, detection device, computer equipment and computer readable storage medium |
-
2018
- 2018-07-20 CN CN201810810454.3A patent/CN109063761B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160241733A1 (en) * | 2015-02-18 | 2016-08-18 | Xerox Corporation | Led illumination uniformity |
| CN107396094A (en) * | 2017-08-17 | 2017-11-24 | 上海大学 | The automatic testing method of single camera damage towards in multi-cam monitoring system |
| CN108198175A (en) * | 2017-12-28 | 2018-06-22 | 广东欧珀移动通信有限公司 | Detection method, detection device, computer equipment and computer readable storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111310578A (en) * | 2020-01-17 | 2020-06-19 | 上海优加利健康管理有限公司 | Method and device for generating heart beat data sample classification network |
| CN111310578B (en) * | 2020-01-17 | 2023-05-02 | 上海乐普云智科技股份有限公司 | Method and device for generating heart beat data sample classification network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109063761B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6629513B2 (en) | Liveness inspection method and apparatus, and video processing method and apparatus | |
| US8792722B2 (en) | Hand gesture detection | |
| US8750573B2 (en) | Hand gesture detection | |
| JP6488647B2 (en) | Object tracking device, object tracking system, object tracking method, display control device, object detection device, program, and recording medium | |
| Wu et al. | One shot learning gesture recognition from rgbd images | |
| CN104915673B (en) | A kind of objective classification method and system of view-based access control model bag of words | |
| KR102257897B1 (en) | Apparatus and method for liveness test,and apparatus and method for image processing | |
| US20200380319A1 (en) | System and method for facilitating graphic-recognition training of a recognition model | |
| CN108960278A (en) | Use the novetly detection of the discriminator of production confrontation network | |
| CN108416902A (en) | Real-time object identification method based on difference identification and device | |
| CN108875522A (en) | Face cluster methods, devices and systems and storage medium | |
| CN106663126A (en) | Video processing for motor task analysis | |
| WO2020134102A1 (en) | Article recognition method and device, vending system, and storage medium | |
| US9501719B1 (en) | System and method for verification of three-dimensional (3D) object | |
| CN109460733A (en) | Recognition of face in-vivo detection method and system based on single camera, storage medium | |
| CN106131405A (en) | Image processing method and device and for its image capturing system | |
| CN108875509A (en) | Biopsy method, device and system and storage medium | |
| CN108229375B (en) | Method and device for detecting face image | |
| JP2020526835A (en) | Devices and methods that dynamically identify a user's account for posting images | |
| KR20190018274A (en) | Method and apparatus for recognizing a subject existed in an image based on temporal movement or spatial movement of a feature point of the image | |
| CN109410138A (en) | Modify jowled methods, devices and systems | |
| WO2015064292A1 (en) | Image feature amount-related processing system, processing method, and program | |
| CN110490058A (en) | Training method, device, system and the computer-readable medium of pedestrian detection model | |
| CN109063761A (en) | Diffuser dropping detection method, device and electronic equipment | |
| US20060257017A1 (en) | Classification methods, classifier determination methods, classifiers, classifier determination devices, and articles of manufacture |
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 |