CN109840501A - A kind of image processing method and device, electronic equipment, storage medium - Google Patents
A kind of image processing method and device, electronic equipment, storage medium Download PDFInfo
- Publication number
- CN109840501A CN109840501A CN201910098425.3A CN201910098425A CN109840501A CN 109840501 A CN109840501 A CN 109840501A CN 201910098425 A CN201910098425 A CN 201910098425A CN 109840501 A CN109840501 A CN 109840501A
- Authority
- CN
- China
- Prior art keywords
- image
- image processing
- group
- controller
- hyper parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The embodiment of the present disclosure discloses a kind of image processing method, this method comprises: S101, i-th group of hyper parameter of acquisition and the i-th controller；S102, i-th group of hyper parameter is adjusted by the i-th controller, generates i+1 group hyper parameter；S103, training image processing model is treated based on i+1 group hyper parameter and sample image be trained test, the corresponding i+1 image measurement result of acquisition i+1 image processing model；S104, when determining that i+1 image measurement result is unsatisfactory for default effect condition, according to i+1 image measurement result the i-th controller of training, acquisition i+1 controller；S105, it enables i be equal to i+1, goes to step S102, until obtaining the corresponding m-th image measurement result of m-th image processing model and meeting default effect condition；S106, m-th image processing model is determined as to optimum image processing model.By implementing above scheme, hyper parameter regulated efficiency and image processing effect are improved.
Description
Technical field
This disclosure relates to technical field of image processing more particularly to a kind of image processing method and device, electronic equipment, deposit
Storage media.
Background technique
Currently, the models such as machine learning and deep learning are widely used to computer vision, natural language processing, recommendation
The numerous areas such as system.For example, image processing model can identify the image got in image processing process, specifically
, in automatic Pilot application field, image processing model can go out barrier therein from the road conditions image recognition currently obtained,
In monitoring security protection application field, image processing model can identify different faces from video monitoring image, and at image
It manages in model, generally comprises many hyper parameters, the final effect of determination and model to model has large effect.
In the prior art, mainly adjust hyper parameter by following three kinds of methods: first method is to manually adjust, and is passed through
Hyper parameter adjustment manually is carried out to model, with training pattern, the modelling effect obtained according to training and experience adjust super ginseng repeatedly
Number, until obtaining the preferable model of effect, second method is force search, specifies limited hyper parameter space, writes specific
Program wherein each group of hyper parameter is scanned for, obtain one group of optimal hyper parameter, the third method is genetic algorithm, i.e.,
A preferable hyper parameter space is evolved into, and then obtain by duplicate selection, heredity and variation in initial hyper parameter space
Obtain preferably one group of hyper parameter.
However, above-mentioned three kinds of methods are to search for one by one to determine the preferable hyper parameter of effect, not only image procossing is used
Image processing model adjustment hyper parameter efficiency it is lower, and hyper parameter adjusted may and it is non-optimal, according to adjustment
The treatment effect that the image processing model that hyper parameter afterwards is determined carries out image procossing is also bad.
Summary of the invention
The embodiment of the present disclosure is intended to provide a kind of image processing method and device, electronic equipment, storage medium, can be improved
The efficiency and image processing effect that hyper parameter adjusts in image processing model.
The technical solution of the embodiment of the present disclosure is achieved in that
The embodiment of the present disclosure provides a kind of image processing method, comprising:
S101, i-th group of hyper parameter and the i-th controller are obtained；I is the natural number more than or equal to 1；
S102, i-th group of hyper parameter is adjusted by i-th controller, generates i+1 group hyper parameter；
S103, treated based on the i+1 group hyper parameter and sample image training image processing model be trained test,
Obtain the corresponding i+1 image measurement result of i+1 image processing model；
S104, when determining that the i+1 image measurement result is unsatisfactory for default effect condition, according to the i+1
A image measurement result training i-th controller, obtains i+1 controller；
S105, it enables i be equal to i+1, goes to step S102, until obtaining the corresponding m-th image of m-th image processing model
Until test result meets the default effect condition；M is the natural number greater than i+1；
S106, the m-th image processing model is determined as to optimum image processing model.
It is described that i-th group of hyper parameter is adjusted by i-th controller in above-mentioned image processing method,
Generate i+1 group hyper parameter, comprising:
Hyper parameter in i-th group of hyper parameter is combined, i-th group of pretreatment parameter is obtained；
I-th group of pretreatment parameter is inputted into i-th controller, generates i+1 group parameter to be processed；
The i+1 group parameter to be processed is split, the i+1 group hyper parameter is obtained.
In above-mentioned image processing method, the sample image includes training image and test image, described based on described
I+1 group hyper parameter and sample image treat training image processing model and are trained test, obtain i+1 image procossing mould
The corresponding i+1 image measurement result of type, comprising:
The i+1 group hyper parameter is substituted into described to obtain object module in training image processing model；
The object module is trained according to the training image, obtains the i+1 image processing model；
Measure of merit is carried out to the i+1 image processing model according to the test image, obtains the i+1
Image measurement result.
It is described that training image is treated based on the i+1 group hyper parameter and sample image in above-mentioned image processing method
Processing model is trained test, after obtaining the corresponding i+1 image measurement result of i+1 image processing model, institute
State method further include:
When determining that the i+1 image measurement result meets the default effect condition, by the i+1 image
Processing model is determined as the optimum image processing model.
It is described according to i+1 image measurement result training i-th control in above-mentioned image processing method
Device obtains i+1 controller, comprising:
Obtain i-th of image measurement result；
When determining that the i+1 image measurement result is higher than i-th of image measurement result, Xiang Suoshu i-th is controlled
Device processed sends positive feedback signal to carry out controller training, obtains the i+1 controller；
When determining the i+1 image measurement result lower than i-th of image measurement result, Xiang Suoshu i-th is controlled
Device processed sends negative-feedback signal to carry out controller training, obtains the i+1 controller.
In above-mentioned image processing method, i be equal to 1 when, it is described by i-th controller to i-th group of hyper parameter
It is adjusted, before generating i+1 group hyper parameter, the method also includes:
Obtain default hyper parameter group；
The default hyper parameter group is determined as the 1st group of hyper parameter；
Survey is trained to training image processing model to described based on the 1st group of hyper parameter and the sample image
Examination obtains the corresponding 1st image measurement result of the 1st image processing model；
When determining that the 1st model test results are unsatisfactory for the default effect condition, prescription controller is obtained；
The prescription controller is determined as the 1st controller.
The embodiment of the present disclosure additionally provides a kind of image processing method, the method also includes:
Image to be processed input optimum image is handled into model；
Identifying processing is carried out to the image to be processed according to the optimum image model, obtains the image pair to be processed
The image recognition result answered.
The embodiment of the present disclosure provides a kind of image processing apparatus, and described image processing unit includes:
Module is obtained, for obtaining i-th group of hyper parameter and the i-th controller；I is the natural number more than or equal to 1；
Test module is adjusted, for being adjusted by i-th controller to i-th group of hyper parameter, generates i+1
Group hyper parameter；Training image processing model is treated based on the i+1 group hyper parameter and sample image and is trained test, is obtained
The corresponding i+1 image measurement result of i+1 image processing model；
Judgment module, when for determining that the i+1 image measurement result is unsatisfactory for default effect condition, according to institute
I+1 image measurement result training i-th controller is stated, i+1 controller is obtained；It enables i be equal to i+1, goes to the tune
Whole test module, until obtaining the corresponding m-th image measurement result of m-th image processing model meets the default effect item
Until part；Wherein, M is the natural number greater than i+1；
Model determining module, for the m-th image processing model to be determined as optimum image processing model.
In above-mentioned image processing apparatus, the adjustment test module, specifically for that will surpass in i-th group of hyper parameter
Parameter is combined, and obtains i-th group of pretreatment parameter；I-th group of pretreatment parameter is inputted into i-th controller, is generated
I+1 group parameter to be processed；Parameter to be processed after the i+1 group is split, the i+1 group hyper parameter is obtained.
In above-mentioned image processing apparatus, the sample image includes training image and test image, the adjustment test
Module is specifically used for substituting into the i+1 group hyper parameter described to obtain object module in training image processing model；Root
The object module is trained according to the training image, obtains the i+1 image processing model；According to the test
Image carries out measure of merit to the i+1 image processing model, obtains the i+1 image measurement result.
In above-mentioned image processing apparatus, the model determining module is also used to determine the i+1 image measurement
When as a result meeting the default effect condition, the i+1 image processing model is determined as the optimum image and handles mould
Type.
In above-mentioned image processing apparatus, the adjustment test module is specifically used for obtaining i-th of image measurement result；
Determine that the i+1 image measurement result is higher than i-th of image measurement as a result, sending just to i-th controller
Feedback signal obtains the i+1 controller to carry out controller training；Determine that the i+1 image measurement result is low
When i-th of image measurement result, the i-th controller of Xiang Suoshu sends negative-feedback signal to carry out controller training, obtains
The i+1 controller.
In above-mentioned image processing apparatus, when i is equal to 1,
The acquisition module is also used to obtain default hyper parameter group；The default hyper parameter group is determined as the 1st group of super ginseng
Number；
The adjustment test module is also used to based on the 1st group of hyper parameter and the sample image to described wait train
Image processing model is trained test, obtains the corresponding 1st image measurement result of the 1st image processing model；
The acquisition module is also used to determine that the 1st image measurement result is unsatisfactory for the default effect condition
When, obtain prescription controller；The prescription controller is determined as the 1st controller.
The embodiment of the present disclosure provides a kind of image processing apparatus, and described image processing unit includes:
Input module, for image to be processed input optimum image to be handled model；
Processing module carries out identifying processing to the image to be processed for handling model according to the optimum image, obtains
Obtain the corresponding image recognition result of the image to be processed.
The embodiment of the present disclosure provides a kind of electronic equipment, and the electronic equipment includes: that processor, memory and communication are total
Line；Wherein,
The communication bus, for realizing the connection communication between the processor and the memory；
The processor determines program for executing the model stored in the memory, to realize above-mentioned image procossing
Method.
The embodiment of the present disclosure provides a kind of computer readable storage medium, and the computer-readable recording medium storage has
One or more program, one or more of programs can be executed by one or more processor, above-mentioned to realize
Image processing method.
It can be seen that S101, obtaining i-th group of hyper parameter and the i-th controller in the technical solution of the embodiment of the present disclosure；i
For the natural number more than or equal to 1；S102, i-th group of hyper parameter is adjusted by the i-th controller, generates the super ginseng of i+1 group
Number；S103, training image processing model is treated based on i+1 group hyper parameter and sample image be trained test, acquisition i+1
The corresponding i+1 image measurement result of a image processing model；S104, determine that i+1 image measurement result is unsatisfactory for
When default effect condition, according to i+1 image measurement result the i-th controller of training, i+1 controller is obtained；S105, i is enabled
Equal to i+1, step S102 is gone to, is preset until obtaining the corresponding m-th image measurement result of m-th image processing model and meeting
Until effect condition；Wherein, M is the natural number greater than i+1；M-th image processing model is determined as optimum image processing mould
Type.That is, the technical solution that the embodiment of the present disclosure provides, based on sample image and one group of hyper parameter training and is tested wait instruct
Practice image processing model, to obtain image measurement as a result, effective to carry out automatically to hyper parameter according to image measurement result
Adjustment, without being scanned for one by one to adjust hyper parameter to hyper parameter space, to not only increase image processing model
The efficiency of middle hyper parameter adjustment, also improves image processing effect.
Detailed description of the invention
Fig. 1 is a kind of flow diagram one for image processing method that the embodiment of the present disclosure provides；
Fig. 2 is a kind of illustrative preprocessing process schematic diagram that the embodiment of the present disclosure provides；
Fig. 3 is the schematic diagram that a kind of illustrative model that the embodiment of the present disclosure provides determines detailed process；
Fig. 4 is a kind of flow diagram two for image processing method that the embodiment of the present disclosure provides；
Fig. 5 is a kind of structural schematic diagram one for image processing apparatus that the embodiment of the present disclosure provides；
Fig. 6 is a kind of structural schematic diagram two for image processing apparatus that the embodiment of the present disclosure provides；
Fig. 7 is the structural schematic diagram for a kind of electronic equipment that the embodiment of the present disclosure provides.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present disclosure, the technical solution in the embodiment of the present disclosure is carried out clear, complete
Site preparation description.
Present disclose provides a kind of image processing methods.Fig. 1 is a kind of image processing method that the embodiment of the present disclosure provides
Flow diagram one.As shown in Figure 1, mainly comprising the steps that
S101, i-th group of hyper parameter and the i-th controller are obtained；I is the natural number more than or equal to 1.
In embodiment of the disclosure, image processing apparatus needs first to get i-th group of hyper parameter and the i-th controller.
It should be noted that in embodiment of the disclosure, i is the natural number more than or equal to 1, specific this public affairs of value of i
Embodiment is opened to be not construed as limiting.
It should be noted that in embodiment of the disclosure, i-th group of hyper parameter may include multiple hyper parameters, specifically
I-th group of hyper parameter and the i-th controller embodiment of the present disclosure are not construed as limiting.
S102, i-th group of hyper parameter is adjusted by the i-th controller, generates i+1 group hyper parameter.
In embodiment of the disclosure, image processing apparatus can adjust i-th group of hyper parameter by the i-th controller
It is whole, to generate i+1 group hyper parameter.
It should be noted that in embodiment of the disclosure, when i is equal to 1, image processing apparatus passes through the i-th controller pair
I-th group of hyper parameter is adjusted, before generation i+1 group hyper parameter, further includes: obtains default hyper parameter group；It will default super ginseng
Array is determined as the 1st group of hyper parameter；Training image processing model, which is treated, based on the 1st group of hyper parameter and sample image is trained survey
Examination obtains the corresponding 1st image measurement result of the 1st image processing model；Determine that the 1st image measurement result is unsatisfactory for
When default effect condition, prescription controller is obtained；Prescription controller is determined as the 1st controller.
It should be noted that being stored with default effect condition in embodiment of the disclosure, in image processing apparatus, this is pre-
If effect condition can be set according to actual needs by user, specifically the default effect condition embodiment of the present disclosure is not limited
It is fixed.
It is understood that in embodiment of the disclosure, practical the 1st group of hyper parameter is one group of initial hyper parameter, figure
Model is handled to training image as processing unit can directly substitute into the 1st group of hyper parameter, to be trained test, obtains the 1st
Image measurement shows the 1st image procossing mould as a result, when determining that the 1st image measurement result is unsatisfactory for default effect condition
The image processing effect of type is bad, therefore, obtains the 1st controller, further to be adjusted to the 1st group of hyper parameter.It needs
Bright, in embodiment of the disclosure, image processing apparatus is continuous repetition adjustment hyper parameter, to train to training image
Manage model.
It should be noted that in embodiment of the disclosure, equally may include multiple super ginsengs in i+1 group hyper parameter
Number, the specific hyper parameter quantity embodiment of the present disclosure are not construed as limiting.
Specifically, in embodiment of the disclosure, image processing apparatus carries out i-th group of hyper parameter by the i-th controller
Adjustment generates i+1 group hyper parameter, comprising: is combined the hyper parameter in i-th group of hyper parameter, obtains i-th group of pretreatment ginseng
Number；I-th group of pretreatment parameter is inputted into the i-th controller, generates i+1 group parameter to be processed；By i+1 group parameter to be processed into
Row is split, and obtains i+1 group hyper parameter.
It should be noted that in embodiment of the disclosure, it can be with the super of two or more in i-th group of hyper parameter
Parameter, image processing apparatus pre-process the association that can reinforce between certain hyper parameters to i-th group of hyper parameter, consequently facilitating
I-th controller generates the better one group of new hyper parameter of effect, i.e. i+1 group hyper parameter.
Fig. 2 is a kind of illustrative preprocessing process schematic diagram that the embodiment of the present disclosure provides.As shown in Fig. 2, i-th group super
Parameter includes: hyper parameter P1-P8, and image processing apparatus can be according to default pretreatment mode by the super ginseng in i-th group of hyper parameter
Number is combined, specifically, first combining hyper parameter P1 and hyper parameter P2 according to default pretreatment mode, is obtained and is combined result
Hyper parameter P3, hyper parameter P4 and hyper parameter P5 are combined by H1, are obtained and are combined result H2, by hyper parameter P6 and hyper parameter P7
It is combined, obtains and combine result H3, and then will be combined in conjunction with result H2 and in conjunction with result H3, obtain and combine result
H4, finally, will be combined in conjunction with result H1, in conjunction with result H4 and hyper parameter P8, to obtain i-th group of pretreatment parameter, i.e.,
It can be input in the i-th controller, and default post processing mode and the default pretreatment mode are exactly the opposite, are inverse process, also
It is to be split to the i+1 group parameter to be processed of the i-th controller output, and then obtain i+1 group hyper parameter, it is no longer superfluous herein
It states.
It should be noted that in embodiment of the disclosure, image processing apparatus can carry out hyper parameter in different ways
Fractionation and combination, specific mode can preset according to actual needs by user, and the embodiment of the present disclosure is not construed as limiting.
It is understood that in embodiment of the disclosure, i-th group of hyper parameter is adjusted by the i-th controller,
Exactly i-th group of hyper parameter is optimized, so that compared to the i-th group hyper parameter of i+1 group hyper parameter generated is more suitable for wait instruct
Practice image processing model.
S103, treated based on i+1 group hyper parameter and sample image training image processing model be trained test, acquisition
The corresponding i+1 image measurement result of i+1 image processing model.
In embodiment of the disclosure, image processing apparatus can be based on i+1 after generating i+1 group hyper parameter
Group hyper parameter treats training image processing model and is trained test, to obtain i+1 image processing model corresponding i-th
+ 1 image measurement result.
It should be noted that can be not instruct also arbitrarily to training image processing model in embodiment of the disclosure
Practice into the machine learning initial model or deep learning initial model for having image-capable, specifically to training image processing
The model embodiment of the present disclosure is not construed as limiting.
Specifically, in embodiment of the disclosure, sample image includes: training image and test image, image procossing dress
It sets and training image processing model is treated based on i+1 group hyper parameter is trained test, acquisition i+1 image processing model pair
The i+1 image measurement result answered, comprising: substitute into i+1 group hyper parameter to obtain target in training image processing model
Model；Object module is trained according to training image, obtains i+1 image processing model；According to test image to i-th
+ 1 image processing model carries out measure of merit, obtains i+1 image measurement result.
It should be noted that in embodiment of the disclosure, training image and test chart are stored in image processing apparatus
Picture, for being trained test to model, specific training image and test image can be independently arranged by user, and the disclosure is real
Example is applied to be not construed as limiting.
It is understood that i+1 group hyper parameter is updated to wait train by image processing apparatus in the implementation of the disclosure
In image processing model, object module is obtained, in object module, there is also need to be trained the uncertain ginseng that just can determine that
Number, therefore, image processing apparatus are further trained object module according to default training set, scheme so that it is determined that going out to training
As processing model in institute setting in need parameter, that is, obtain i+1 image processing model.
Illustratively, in embodiment of the disclosure, the i+1 image processing model that image processing apparatus trains is
Specifically in identification image personage region and inhuman object area, therefore, image processing apparatus is according to default training set, i.e., one
The image of fixed number amount carries out measure of merit to i+1 image processing model, correctly identifies personage region and non-to obtain
The accuracy rate in personage region is A, and accuracy rate A is determined as the corresponding i+1 image measurement knot of i+1 image processing model
Fruit.
It should be noted that in embodiment of the disclosure, due to that can be different type to training image processing model
Image processing model, and different types of image processing model its effect has differences, therefore, evaluation model test result
Mode also different from needs the effect implemented according to model to be assessed, can specifically preset corresponding effect condition,
The embodiment of the present disclosure is not construed as limiting.
S104, when determining that i+1 image measurement result is unsatisfactory for default effect condition, according to the survey of i+1 image
Test result trains the i-th controller, obtains i+1 controller.
In embodiment of the disclosure, image processing apparatus is after obtaining i+1 image measurement result, determine i-th+
Whether 1 image measurement result meets default effect condition, is determining that i+1 image measurement result is unsatisfactory for default effect
When condition, according to i+1 image measurement result the i-th controller of training, i+1 controller is obtained.
Illustratively, in embodiment of the disclosure, the i+1 image processing model tool that image processing apparatus trains
Body the personage region in image and inhuman object area for identification, obtain i-th of image measurement as a result, correctly identifying personage
The accuracy rate A in region and inhuman object area, presetting effect condition is that accuracy rate is more than or equal to B, if A is less than B, image processing apparatus
Determine that i+1 image measurement result is unsatisfactory for default effect condition, correspondingly, if A is more than or equal to B, image processing apparatus
Determine that i+1 image measurement result meets default effect condition.
It is understood that in embodiment of the disclosure, image processing apparatus handles model to training image in training
During, the setting of hyper parameter has large effect to image processing model effect, therefore, if train before i-th+
The corresponding i+1 image measurement result of 1 image processing model is unsatisfactory for default effect condition, that is, the i+1 trained
The modelling effect of image processing model is poor, needs further to adjust the i+1 for training i+1 image processing model
Group hyper parameter, and to adjust i+1 group hyper parameter, it is necessary to the i-th controller of training generation i+1 group hyper parameter first, thus
Obtain i+1 controller.
It should be noted that in embodiment of the disclosure, image processing apparatus is instructed according to i+1 image measurement result
Practice the i-th controller, is specifically adjusted in the i-th controller according to i+1 image measurement result for one group of hyper parameter to input
It is adjusted, to export the algorithm or rule of one group of new hyper parameter, alternatively, being the relevant parameter of algorithm or rule, at image
Reason device can be adjusted according to certain default adjustment law, and the embodiment of the present disclosure is not construed as limiting.
Specifically, in embodiment of the disclosure, image processing apparatus is according to i+1 model test results training i-th
Controller obtains i+1 controller, comprising: obtains i-th of image measurement result；Determine that i+1 image measurement result is high
When i-th of image measurement result, positive feedback signal is sent to the i-th controller to carry out controller training, obtains i+1 control
Device；When determining i+1 image measurement result lower than i-th of image measurement result, negative-feedback signal is sent to the i-th controller
To carry out controller training, i+1 controller is obtained.
It is understood that in embodiment of the disclosure, image processing apparatus before generating i+1 group hyper parameter,
It has been based on i-th group of hyper parameter and treats training image processing model and be trained test, obtain i-th of image processing model correspondence
I-th of image measurement as a result, and i-th of image measurement needs to continue mould the result is that be unsatisfactory for default effect condition
Type training, therefore, image processing apparatus can after obtaining i+1 image measurement result, before being directly obtained i-th
A image measurement as a result, to compare two image measurements as a result, specific i-th of image measurement result embodiment of the present disclosure not
It limits.
It is understood that in embodiment of the disclosure, image processing apparatus compare i+1 image measurement result and
I-th of image measurement as a result, can determine training the i-th controller direction, i.e., referred to positive feedback signal or negative-feedback signal
Show that the i-th controller adjusts the direction of internal relevant parameter.
Illustratively, in embodiment of the disclosure, image processing apparatus trains i+1 image processing model and
I-th of image processing model is specifically used for personage region and inhuman object area in identification image, is based on i-th of image procossing mould
What type obtained correctly identifies that i-th of image measurement result of personage region and inhuman object area is accuracy rate A1, correspondingly, the
I+1 image measurement result is accuracy rate A2, and default effect condition is more than or equal to A3, A1 and A2 for accuracy rate and is respectively less than A3, if
A1 is greater than A2, and image processing apparatus sends positive feedback signal to the i-th controller to carry out controller training, specifically, the i-th control
Device increases internal relevant parameter according to default increase mode according to positive feedback signal, if A1 is less than or equal to A2, image processing apparatus
Negative-feedback signal is sent to carry out controller training to the i-th controller, specifically, the i-th controller can be with according to negative-feedback signal
Default reduction mode reduces internal relevant parameter.
It should be noted that in embodiment of the disclosure, image processing apparatus is based on i+1 group hyper parameter and treats training
Image processing model is trained test, and after obtaining i+1 image measurement result, i+1 image measurement result may
Meet default effect condition, therefore, after step s 103, determines that i+1 image measurement result meets default effect item
When part, i+1 image processing model is determined as optimum image processing model by image processing apparatus.
It is understood that in embodiment of the disclosure, the final purpose of image processing apparatus is as determined optimal
Image processing model, therefore, if i+1 image measurement result meets default effect condition, image processing apparatus also will at this time
The process for determining model is terminated, also, the hyper parameter in the model of optimum image processing at this time, i.e. i+1 group hyper parameter are most
One group of excellent hyper parameter.
S105, it enables i be equal to i+1, goes to step S102, until obtaining the corresponding m-th image of m-th image processing model
Until test result meets default effect condition；Wherein, M is the natural number greater than i+1.
In embodiment of the disclosure, image processing apparatus enables i be equal to i+1, goes to after obtaining i+1 controller
Step S102, until the corresponding default effect condition of m-th image measurement result satisfaction of acquisition m-th image processing model is
Only.
It is understood that in embodiment of the disclosure, image processing apparatus is after obtaining i+1 controller, i.e.,
It enables i be equal to i+1, goes to step S102, i+1 group hyper parameter is adjusted by i+1 controller, generate the super ginseng of i+2 group
Number, later, continues to execute step S103: treating training image processing model based on the i-th+2 group hyper parameter and is trained test, obtains
Corresponding the i-th+2 image measurements of the i-th+2 image processing models are obtained as a result, simultaneously further judging the i-th+2 model measurement knots
Whether fruit meets default effect condition, if not satisfied, executing step S104: determining that the i-th+2 image measurement results are discontented
When foot presets effect condition, according to the i-th+2 image measurement results training i+1 controllers, the i-th+2 controller is obtained,
And then i is enabled to be equal to i+1, step S102 is gone to, identical step is repeated, until m-th image processing model is obtained,
And the corresponding m-th image measurement result of m-th image processing model meets default effect condition, that is to say, that will adjust every time
Hyper parameter group after whole is as the input adjusted next time, so that constantly circulation adjustment hyper parameter group, final to obtain the super ginseng of M group
Number, and training image processing model is treated based on M group hyper parameter and is trained test, the m-th iconic model of acquisition is corresponding
M-th image measurement the result is that meeting default effect condition.
S106, m-th image processing model is determined as to optimum image processing model.
In embodiment of the disclosure, image processing apparatus is obtaining m-th image processing model, and at m-th image
Managing the corresponding default effect condition of m-th image measurement result satisfaction of model can be by M after terminating model determination process
A image processing model is determined as optimum image processing model, to handle model realization image processing application by optimum image.
It should be noted that in embodiment of the disclosure, the corresponding m-th image measurement of m-th image processing model
As a result meet default effect condition, also indicate that the image processing effect of m-th image processing model is optimal, it is therefore, corresponding
M-th image processing model can be determined as optimum image processing model, correspondingly, in m-th image processing model
M group hyper parameter can also be determined as one group of optimal hyper parameter.
Fig. 3 is a kind of schematic diagram for illustrative model training detailed process that the embodiment of the present disclosure provides.Such as Fig. 3 institute
Show, input the i-th controller after i-th group of hyper parameter is combined, then obtain the after splitting from the parameter that the i-th controller exports
I+1 group hyper parameter is inputted training test environment, to be based on i+1 group hyper parameter and sample later by i+1 group hyper parameter
Image treats training image processing model and is trained test, obtains the corresponding i+1 image of i+1 image processing model
Test result when determining that i-th of image measurement result is unsatisfactory for default effect condition, then compares the survey of i+1 image later
Test result and i-th of image measurement are as a result, send positive feedback signal or negative-feedback signal, automatic the i-th control of training to the i-th controller
Device processed is to generate i+1 controller, after the process, enable i be equal to i+1, repeat before the step of, until obtain M
The corresponding m-th image measurement result of a image processing model meets default effect condition.
The embodiment of the present disclosure provides a kind of image processing method, S101, obtains i-th group of hyper parameter and the i-th controller；i
For the natural number more than or equal to 1；S102, i-th group of hyper parameter is adjusted by the i-th controller, generates the super ginseng of i+1 group
Number；S103, training image processing model is treated based on i+1 group hyper parameter and sample image be trained test, acquisition i+1
The corresponding i+1 image measurement result of a image processing model；S104, determine that i+1 image measurement result is unsatisfactory for
When default effect condition, according to i+1 image measurement result the i-th controller of training, i+1 controller is obtained；S105, i is enabled
Equal to i+1, step S102 is gone to, is preset until obtaining the corresponding m-th image measurement result of m-th image processing model and meeting
Until effect condition；Wherein, M is the natural number greater than i+1；S106, m-th image processing model is determined as at optimum image
Manage model.That is, the technical solution that the embodiment of the present disclosure provides, based on sample image and one group of hyper parameter training and test
Model is handled to training image, to obtain image measurement as a result, to have automatically to hyper parameter according to image measurement result
The adjustment of effect, without being scanned for one by one to adjust hyper parameter to hyper parameter space, to not only increase image procossing
The efficiency that hyper parameter adjusts in model, also improves image processing effect.
Another embodiment of the disclosure provides a kind of image processing method, and Fig. 4 is a kind of figure that the embodiment of the present disclosure provides
As the flow diagram two of processing method.As shown in figure 4, mainly comprising the steps that
S401, image to be processed input optimum image is handled into model.
In embodiment of the disclosure, image processing apparatus can be by image to be treated, i.e., image input to be processed
The optimum image that above-mentioned image processing apparatus is determined handles model, carries out image procossing.
It should be noted that in embodiment of the disclosure, it is image in a upper embodiment that optimum image, which handles model,
The model that processing unit is determined, the specific optimum image processing model embodiment of the present disclosure are not construed as limiting.
It should be noted that in embodiment of the disclosure, what image to be processed can get for image processing apparatus
Arbitrary image, the specific image embodiment of the present disclosure to be processed are not construed as limiting.
S402, model is handled according to optimum image to image to be processed progress identifying processing, it is corresponding to obtain image to be processed
Image recognition result.
In embodiment of the disclosure, image processing apparatus handles model according to optimum image and knows to image to be processed
Other places reason, obtains the corresponding image recognition result of image to be processed.
It should be noted that in embodiment of the disclosure, image processing apparatus handles model according to optimum image can be with
Identifying processing is carried out to each section image information for including in image to be processed, for example, identifying the image districts such as personage, plant
Domain, further, it is also possible to identify the region for needing to replace in wherein image to be processed, such as background area, the automatic background that carries out is replaced
Change, etc..Specific identification processing procedure and the image recognition result embodiment of the present disclosure are not construed as limiting.
Illustratively, in embodiment of the disclosure, optimum image processing model is the human face recognition model after training, because
This, image processing apparatus will can need to carry out the image of recognition of face, i.e., the recognition of face after image input training to be processed
Model, and human face recognition model carries out recognition of face to the image of input, and the area of face is belonged in the image to identify input
Domain, the i.e. corresponding face recognition result of acquisition input picture.Certainly, image to be processed may be multiple, and image procossing fills
It sets in the human face recognition model by multiple after needing the image of recognition of face to be separately input to training, the recognition of face mould after training
More whether type can not only identify the human face region in every image, can also be further processed, be same
The corresponding facial image of entity object, so that comparison result is exported, and the comparison result is final image recognition result.
It is understood that in embodiment of the disclosure, image to be processed is inputted optimum image by image processing apparatus
Model is handled, model is handled according to optimum image, identifying processing is carried out to image to be processed, is all based on optimum image processing mould
Type for example, extracting characteristics of image, image key points etc., and specifically handles, for example, to image the recognition capability of image
The region of middle characterization different types of information is classified, or picks out the concrete applications such as the preferable image of image effect,
It is based on the cognition identification to image, the embodiment of the present disclosure is not construed as limiting.
The another embodiment of the disclosure provides a kind of image processing apparatus.Fig. 5 is a kind of figure that the embodiment of the present disclosure provides
As the structural schematic diagram one of processing unit.As shown in figure 5, image processing apparatus includes:
Module 501 is obtained, for obtaining i-th group of hyper parameter and the i-th controller；I is the natural number more than or equal to 1；
Test module 502 is adjusted, for being adjusted by i-th controller to i-th group of hyper parameter, generation the
I+1 group hyper parameter；Training image processing model, which is treated, based on the i+1 group hyper parameter and sample image is trained test,
Obtain the corresponding i+1 image measurement result of i+1 image processing model；
Judgment module 503, when for determining that the i+1 image measurement result is unsatisfactory for default effect condition, root
According to i+1 image measurement result training i-th controller, i+1 controller is obtained；It enables i be equal to i+1, goes to institute
State adjustment test module 502, until obtain the corresponding m-th image measurement result of m-th image processing model meet it is described pre-
If until effect condition；Wherein, M is the natural number greater than i+1；
Model determining module 504, for the m-th image processing model to be determined as optimum image processing model.
Optionally, the adjustment test module 502, specifically for tying the hyper parameter in i-th group of hyper parameter
It closes, obtains i-th group of pretreatment parameter；I-th group of pretreatment parameter is inputted into i-th controller, i+1 group is generated and waits locating
Manage parameter；The i+1 group parameter to be processed is split, the i+1 group hyper parameter is obtained.
Optionally, the adjustment test module 502, the sample image include training image and test image, specific use
It is described to obtain object module in training image processing model in substituting into the i+1 group hyper parameter；According to the training figure
As being trained to the object module, the i+1 image processing model is obtained；According to the test image to described
I+1 image processing model carries out measure of merit, obtains the i+1 image measurement result.
Optionally, the model determining module 503 is also used to determine that the i+1 image measurement result meets institute
When stating default effect condition, the i+1 image processing model is determined as the optimum image and handles model.
Optionally, the adjustment test module is specifically used for obtaining i-th of image measurement result；Determine the i+1
When a image measurement result is higher than i-th of image measurement result, the i-th controller of Xiang Suoshu sends positive feedback signal to carry out
Controller training, obtains the i+1 controller；Determine the i+1 image measurement result lower than i-th of image
When test result, the i-th controller of Xiang Suoshu sends negative-feedback signal to carry out controller training, obtains the i+1 controller.
Optionally, when i is equal to 1, the acquisition module 501 is also used to obtain default hyper parameter group；By the default super ginseng
Array is determined as the 1st group of hyper parameter；
The adjustment test module 502 is also used to based on the 1st group of hyper parameter and the sample image to described wait instruct
Practice image processing model and be trained test, obtains the corresponding 1st image measurement result of the 1st image processing model；
The acquisition module 501 is also used to determine that the 1st image measurement result is unsatisfactory for the default effect article
When part, prescription controller is obtained；The prescription controller is determined as the 1st controller.
The embodiment of the present disclosure provides a kind of image processing apparatus, S101, obtains i-th group of hyper parameter and the i-th controller；i
For the natural number more than or equal to 1；S102, i-th group of hyper parameter is adjusted by the i-th controller, generates the super ginseng of i+1 group
Number；S103, training image processing model is treated based on i+1 group hyper parameter and sample image be trained test, acquisition i+1
The corresponding i+1 image measurement result of a image processing model；S104, determine that i+1 image measurement result is unsatisfactory for
When default effect condition, according to i+1 image measurement result the i-th controller of training, i+1 controller is obtained；S105, i is enabled
Equal to i+1, step S102 is gone to, is preset until obtaining the corresponding m-th image measurement result of m-th image processing model and meeting
Until effect condition；Wherein, M is the natural number greater than i+1；S106, m-th image processing model is determined as at optimum image
Manage model.That is, the image processing apparatus that the embodiment of the present disclosure provides, based on sample image and the training of one group of hyper parameter and
Test to training image handle model, thus obtain image measurement as a result, with according to image measurement result automatically to hyper parameter into
The effective adjustment of row, without being scanned for one by one to adjust hyper parameter to hyper parameter space, to not only increase image
The efficiency for handling hyper parameter adjustment in model, also improves image processing effect.
The another embodiment of the disclosure provides a kind of image processing apparatus, and Fig. 6 is a kind of figure that the embodiment of the present disclosure provides
As the structural schematic diagram two of processing unit.As shown in fig. 6, described image processing unit includes:
Input module 601, for image to be processed input optimum image to be handled model；
Processing module 602 carries out identifying processing to the image to be processed for handling model according to the optimum image,
Obtain the corresponding image recognition result of the image to be processed.
The embodiment of the present disclosure provides a kind of electronic equipment.Fig. 7 is a kind of electronic equipment that the embodiment of the present disclosure provides
Structural schematic diagram.As shown in fig. 7, electronic equipment includes: processor 701, memory 702 and communication bus 703；Wherein,
The communication bus 703, for realizing the connection communication between the processor 701 and the memory 702；
The processor 701, for executing the image processing program stored in the memory 702, to realize above-mentioned figure
As processing method.
The embodiment of the present disclosure additionally provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is one or more program, one or more of programs can be executed by one or more processor, on realizing
State image processing method.It is volatile memory (volatile memory) that computer readable storage medium, which can be, such as with
Machine accesses memory (Random-Access Memory, RAM)；Or nonvolatile memory (non-volatile
Memory), such as read-only memory (Read-Only Memory, ROM), flash memory (flash memory), hard disk
(Hard Disk Drive, HDD) or solid state hard disk (Solid-State Drive, SSD)；It is also possible to include above-mentioned memory
One of or any combination respective equipment, such as mobile phone, computer, tablet device, personal digital assistant.
It should be understood by those skilled in the art that, embodiment of the disclosure can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the disclosure
Formula.Moreover, the disclosure, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The disclosure is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present disclosure
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable signal processing equipments to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable signal processing equipments execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable signal processing equipments with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions can also be loaded into computer or other programmable signal processing equipments, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
The above, the only preferred embodiment of the disclosure, are not intended to limit the protection scope of the disclosure.
Claims (10)
1. a kind of image processing method, which is characterized in that the described method includes:
S101, i-th group of hyper parameter and the i-th controller are obtained；I is the natural number more than or equal to 1；
S102, i-th group of hyper parameter is adjusted by i-th controller, generates i+1 group hyper parameter；
S103, treated based on the i+1 group hyper parameter and sample image training image processing model be trained test, acquisition
The corresponding i+1 image measurement result of i+1 image processing model；
S104, when determining that the i+1 image measurement result is unsatisfactory for default effect condition, according to the i+1 figure
As test result training i-th controller, i+1 controller is obtained；
S105, it enables i be equal to i+1, goes to step S102, until obtaining the corresponding m-th image measurement of m-th image processing model
As a result until meeting the default effect condition；M is the natural number greater than i+1；
S106, the m-th image processing model is determined as to optimum image processing model.
2. image processing method according to claim 1, which is characterized in that it is described by i-th controller to described
I-th group of hyper parameter is adjusted, and generates i+1 group hyper parameter, comprising:
Hyper parameter in i-th group of hyper parameter is combined, i-th group of pretreatment parameter is obtained；
I-th group of pretreatment parameter is inputted into i-th controller, generates i+1 group parameter to be processed；
The i+1 group parameter to be processed is split, the i+1 group hyper parameter is obtained.
3. image processing method according to claim 1 or 2, which is characterized in that the sample image includes training image
And test image, it is described training image processing model is treated based on the i+1 group hyper parameter and sample image to be trained survey
Examination obtains the corresponding i+1 image measurement result of i+1 image processing model, comprising:
The i+1 group hyper parameter is substituted into described to obtain object module in training image processing model；
The object module is trained according to the training image, obtains the i+1 image processing model；
Measure of merit is carried out to the i+1 image processing model according to the test image, obtains the i+1 image
Test result.
4. image processing method according to claim 1-3, which is characterized in that described to be based on the i+1 group
Hyper parameter and sample image treat training image processing model and are trained test, and it is corresponding to obtain i+1 image processing model
I+1 image measurement result after, the method also includes:
When determining that the i+1 image measurement result meets the default effect condition, by the i+1 image procossing
Model is determined as the optimum image processing model.
5. image processing method according to claim 1-4, which is characterized in that described according to the i+1
Image measurement result training i-th controller, obtains i+1 controller, comprising:
Obtain i-th of image measurement result；
When determining that the i+1 image measurement result is higher than i-th of image measurement result, the i-th controller of Xiang Suoshu
Positive feedback signal is sent to carry out controller training, obtains the i+1 controller；
When determining the i+1 image measurement result lower than i-th of image measurement result, the i-th controller of Xiang Suoshu
Negative-feedback signal is sent to carry out controller training, obtains the i+1 controller.
6. a kind of image processing method, which is characterized in that the described method includes:
The image to be processed input optimum image that such as any one of claim 1-5 described image processing method determines is handled into mould
Type；
Model is handled according to the optimum image, identifying processing is carried out to the image to be processed, obtain the image pair to be processed
The image recognition result answered.
7. a kind of image processing apparatus, which is characterized in that described image processing unit includes:
Module is obtained, for obtaining i-th group of hyper parameter and the i-th controller；I is the natural number more than or equal to 1；
Test module is adjusted, for being adjusted by i-th controller to i-th group of hyper parameter, it is super to generate i+1 group
Parameter；Training image processing model, which is treated, based on the i+1 group hyper parameter and sample image is trained test, acquisition i-th+
The corresponding i+1 image measurement result of 1 image processing model；
Judgment module, when for determining that the i+1 image measurement result is unsatisfactory for default effect condition, according to described the
I+1 image measurement result training i-th controller, obtains i+1 controller；It enables i be equal to i+1, goes to the adjustment and survey
Die trial block, until obtaining the corresponding m-th image measurement result of m-th image processing model and meeting the default effect condition and be
Only；M is the natural number greater than i+1；
Model determining module, for the m-th image processing model to be determined as optimum image processing model.
8. a kind of image processing apparatus, which is characterized in that described image processing unit includes:
Input module, for image to be processed to be inputted the optimum image that image processing apparatus as claimed in claim 7 determines
Handle model；
Processing module carries out identifying processing to the image to be processed for handling model according to the optimum image, obtains institute
State the corresponding image recognition result of image to be processed.
9. a kind of electronic equipment, which is characterized in that the electronic equipment includes: processor, memory and communication bus；Wherein,
The communication bus, for realizing the connection communication between the processor and the memory；
The processor, for executing the image processing program stored in the memory, to realize any one of claim 1-6
The image processing method.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage have one or
Multiple programs, one or more of programs can be executed by one or more processor, to realize that claim 1-6 appoints
Image processing method described in one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098425.3A CN109840501B (en) | 2019-01-31 | 2019-01-31 | Image processing method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098425.3A CN109840501B (en) | 2019-01-31 | 2019-01-31 | Image processing method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109840501A true CN109840501A (en) | 2019-06-04 |
| CN109840501B CN109840501B (en) | 2021-06-01 |
Family
ID=66884538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910098425.3A Active CN109840501B (en) | 2019-01-31 | 2019-01-31 | Image processing method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109840501B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107004003A (en) * | 2015-11-16 | 2017-08-01 | 华为技术有限公司 | model parameter fusion method and device |
| CN107209873A (en) * | 2015-01-29 | 2017-09-26 | 高通股份有限公司 | Hyper parameter for depth convolutional network is selected |
| CN107590532A (en) * | 2017-08-14 | 2018-01-16 | 华南理工大学 | A kind of hyper parameter dynamic adjusting method based on WGAN |
| US9990687B1 (en) * | 2017-01-19 | 2018-06-05 | Deep Learning Analytics, LLC | Systems and methods for fast and repeatable embedding of high-dimensional data objects using deep learning with power efficient GPU and FPGA-based processing platforms |
| US10049260B1 (en) * | 2017-01-27 | 2018-08-14 | Banuba Limited | Computer systems and computer-implemented methods specialized in processing electronic image data |
| CN108538301A (en) * | 2018-02-13 | 2018-09-14 | 吟飞科技(江苏)有限公司 | A kind of intelligent digital musical instrument based on neural network Audiotechnica |
| CN108718249A (en) * | 2018-04-27 | 2018-10-30 | 广州西麦科技股份有限公司 | Network accelerating method, device based on SDN network and computer readable storage medium |
| CN108805254A (en) * | 2018-04-27 | 2018-11-13 | 中国科学院自动化研究所 | Parameter Optimization System for time series forecasting |
| CN108805268A (en) * | 2018-06-08 | 2018-11-13 | 中国科学技术大学 | Deeply learning strategy network training method based on evolution algorithm |
| US20180357539A1 (en) * | 2017-06-09 | 2018-12-13 | Korea Advanced Institute Of Science And Technology | Electronic apparatus and method for re-learning trained model |
| CN109088785A (en) * | 2018-06-14 | 2018-12-25 | 苏州新研联信息科技有限公司 | A kind of IDC data center apparatus management system |
| CN109213805A (en) * | 2018-09-07 | 2019-01-15 | 东软集团股份有限公司 | A kind of method and device of implementation model optimization |
| US10185803B2 (en) * | 2015-06-15 | 2019-01-22 | Deep Genomics Incorporated | Systems and methods for classifying, prioritizing and interpreting genetic variants and therapies using a deep neural network |
-
2019
- 2019-01-31 CN CN201910098425.3A patent/CN109840501B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107209873A (en) * | 2015-01-29 | 2017-09-26 | 高通股份有限公司 | Hyper parameter for depth convolutional network is selected |
| EP3251058A1 (en) * | 2015-01-29 | 2017-12-06 | Qualcomm Incorporated | Hyper-parameter selection for deep convolutional networks |
| US10185803B2 (en) * | 2015-06-15 | 2019-01-22 | Deep Genomics Incorporated | Systems and methods for classifying, prioritizing and interpreting genetic variants and therapies using a deep neural network |
| CN107004003A (en) * | 2015-11-16 | 2017-08-01 | 华为技术有限公司 | model parameter fusion method and device |
| US9990687B1 (en) * | 2017-01-19 | 2018-06-05 | Deep Learning Analytics, LLC | Systems and methods for fast and repeatable embedding of high-dimensional data objects using deep learning with power efficient GPU and FPGA-based processing platforms |
| US10049260B1 (en) * | 2017-01-27 | 2018-08-14 | Banuba Limited | Computer systems and computer-implemented methods specialized in processing electronic image data |
| US20180357539A1 (en) * | 2017-06-09 | 2018-12-13 | Korea Advanced Institute Of Science And Technology | Electronic apparatus and method for re-learning trained model |
| CN107590532A (en) * | 2017-08-14 | 2018-01-16 | 华南理工大学 | A kind of hyper parameter dynamic adjusting method based on WGAN |
| CN108538301A (en) * | 2018-02-13 | 2018-09-14 | 吟飞科技(江苏)有限公司 | A kind of intelligent digital musical instrument based on neural network Audiotechnica |
| CN108805254A (en) * | 2018-04-27 | 2018-11-13 | 中国科学院自动化研究所 | Parameter Optimization System for time series forecasting |
| CN108718249A (en) * | 2018-04-27 | 2018-10-30 | 广州西麦科技股份有限公司 | Network accelerating method, device based on SDN network and computer readable storage medium |
| CN108805268A (en) * | 2018-06-08 | 2018-11-13 | 中国科学技术大学 | Deeply learning strategy network training method based on evolution algorithm |
| CN109088785A (en) * | 2018-06-14 | 2018-12-25 | 苏州新研联信息科技有限公司 | A kind of IDC data center apparatus management system |
| CN109213805A (en) * | 2018-09-07 | 2019-01-15 | 东软集团股份有限公司 | A kind of method and device of implementation model optimization |
Non-Patent Citations (6)
| Title |
|---|
| CHENYANG SHEN,AT EL.: ""Inetlligent Parameter Tuning in Optimization-based Iterative CT Reconstruction via Deep Reinforcement Learning"", 《ARXIV》 * |
| JAMES BERGSTRA,ET AL.: ""Algorithms for hyper-parameter optimization"", 《NIPS"11：PROCEEDINGS OF THE 24TH INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS》 * |
| LISHA LI,AT EL.: ""Hyperband:Bandit-based Configuration Evaluation for Hyperparameter Optimization"", 《ICLR 2017》 * |
| MARCIN ANDRYCHOWICZ,AT EL.: ""Learning to learn by gradient descent by gradient descent"", 《ARXIV》 * |
| MARSGGBO: ""贝叶斯优化（Bayesian Optimization）深入理解"", 《博客园》 * |
| SAMANTHA HANSEN: ""Using Deep Q-Learning to Control Optimization Hyperparameters"", 《ARXIV》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109840501B (en) | 2021-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101055468A (en) | Robust process model identification in model based control techniques | |
| CN101911272B (en) | Method and apparatus for ultrasonic bonding | |
| CN110223218B (en) | Face image processing method and device, electronic equipment and storage medium | |
| CN104614656B (en) | Semiconductor chip retest system and retest method thereof | |
| CN107346665A (en) | Method, apparatus, equipment and the storage medium of audio detection | |
| US20050049913A1 (en) | Method and apparatus for automated feature selection | |
| CN109858714A (en) | Based on pipe tobacco quality inspection index prediction technique, the apparatus and system for improving neural network | |
| CN109978287A (en) | Polysilicon intelligence production method and system | |
| CN108227664A (en) | Control of product quality equipment and method of quality control based on sample data training | |
| CN106326632A (en) | Object analyzing method and object analyzing system | |
| CN108289634A (en) | Learn the system and method for brain-computer interface for operator | |
| DE102018126429A1 (en) | Processing condition adjustment device and machine learning device | |
| CN116596339B (en) | Feed raw material production adjusting method | |
| CN108985055A (en) | A kind of detection method and system of Malware | |
| CN109858631A (en) | The automaton learning system and method for stream data analysis for concept migration | |
| CN110363355A (en) | A kind of cloud-Bian Xietong the forecast system and method for alumina producing index | |
| van Pelt et al. | The need for integrating neuronal morphology databases and computational environments in exploring neuronal structure and function | |
| CN109840501A (en) | A kind of image processing method and device, electronic equipment, storage medium | |
| CN107526440A (en) | The intelligent electric appliance control method and system of gesture identification based on decision tree classification | |
| CN110032987A (en) | A kind of surface electromyogram signal classification method based on CMAC Neural Network model | |
| CN109961160B (en) | Power grid future operation trend estimation method and system based on tide parameters | |
| CN117399153B (en) | Self-adaptive control method and system for mill | |
| Mannor et al. | Robust approachability and regret minimization in games with partial monitoring | |
| JP2005160020A5 (en) | ||
| Bashiri | A robust desirability-based approach to optimizing multiple correlated responses |
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 |