CN109815969A - A kind of feature extracting method and device based on artificial intelligence image recognition - Google Patents
A kind of feature extracting method and device based on artificial intelligence image recognition Download PDFInfo
- Publication number
- CN109815969A CN109815969A CN201910162758.8A CN201910162758A CN109815969A CN 109815969 A CN109815969 A CN 109815969A CN 201910162758 A CN201910162758 A CN 201910162758A CN 109815969 A CN109815969 A CN 109815969A
- Authority
- CN
- China
- Prior art keywords
- matrix
- feature
- artificial intelligence
- obtains
- characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of feature extracting method and device based on artificial intelligence image recognition, wherein this feature extracting method includes: to obtain fisrt feature matrix, which includes the feature of images to be recognized;Feature extraction is carried out to fisrt feature matrix according to preset convolution algorithm, obtains second characteristic matrix;Recombination is ranked up to second characteristic matrix, obtains third feature matrix;Feature extraction is carried out to third feature matrix according to convolution algorithm, obtains characteristics of image set.Feature extracting method and device based on artificial intelligence image recognition described in the invention, a large amount of and duplicate work in feature extraction operation can be avoided in modeling process, to reduce the workload of establishing of artificial intelligence model, and then it can be improved the efficiency of artificial intelligence model foundation.
Description
Technical field
The present invention relates to artificial intelligence fields, mention in particular to a kind of feature based on artificial intelligence image recognition
Take method and device.
Background technique
Currently, with the fast development of artificial intelligence, the technologies such as image recognition and speech recognition all with fast development,
In, image recognition technology even can be accurately identified the article of most of classifications.However, it has been found in practice that above-mentioned
Image recognition technology be based on artificial intelligence, that is to say, that current image recognition technology be using artificial intelligence model into
Row identification, therefore the degree of difficulty of artificial model's foundation certainly will influence whether the service condition of image recognition technology.According to current
Situation apparently, artificial intelligence model during foundation would generally include feature extraction operation, and this feature extract behaviour
Make that there is a large amount of and duplicate element calling process, it can be seen that workload of the current artificial intelligence model when establishing
It is larger, and then affect the efficiency of artificial intelligence model foundation.
Summary of the invention
In view of the above problems, the present invention provides a kind of feature extracting method and dress based on artificial intelligence image recognition
It sets, a large amount of and duplicate work in feature extraction operation can be avoided in modeling process, to reduce artificial intelligence
Model establishes workload, and then improves the efficiency that artificial intelligence model is established.
To achieve the goals above, the present invention adopts the following technical scheme that:
In a first aspect, the present invention provides a kind of feature extracting methods based on artificial intelligence image recognition, comprising:
Fisrt feature matrix is obtained, the fisrt feature matrix includes the feature of images to be recognized;
Feature extraction is carried out to the fisrt feature matrix according to preset convolution algorithm, obtains second characteristic matrix;
Recombination is ranked up to the second characteristic matrix, obtains third feature matrix;
Feature extraction is carried out to the third feature matrix according to the convolution algorithm, obtains characteristics of image set.
As an alternative embodiment, described carry out spy to the fisrt feature matrix according to preset convolution algorithm
Sign is extracted, and the step of obtaining second characteristic matrix includes:
Data-bias processing is carried out to the fisrt feature matrix according to preset convolution algorithm, obtains side-play amount set;
Offset moment matrix is generated according to the side-play amount set;
Feature extraction is carried out to the fisrt feature matrix according to the offset moment matrix, obtains second characteristic matrix.
As an alternative embodiment, described be ranked up recombination to the second characteristic matrix, third spy is obtained
Levy matrix the step of include:
The processing of matrix transposition is carried out to the second characteristic matrix, obtains third feature matrix.
As an alternative embodiment, described carry out spy to the third feature matrix according to the convolution algorithm
Sign is extracted, and the step of obtaining characteristics of image set includes:
Obtain the convolution kernel that the convolution algorithm includes;
Matrix multiple operation is carried out to the convolution kernel and the third feature matrix, obtains characteristics of image set.
As an alternative embodiment, the method also includes:
Judge whether described image characteristic set meets default feature request;
When described image characteristic set does not meet default feature request, pond Hua Chu is carried out to described image characteristic set
Reason, obtains pond characteristic set;
The pond characteristic set is determined as the fisrt feature matrix, and is executed described according to preset convolution algorithm
The step of is carried out by feature extraction, obtains second characteristic matrix for the fisrt feature matrix.
Second aspect, the present invention provides a kind of feature deriving means based on artificial intelligence image recognition, comprising:
Module is obtained, for obtaining fisrt feature matrix, the fisrt feature matrix includes the feature of images to be recognized;
Conversion module obtains for carrying out feature extraction to the fisrt feature matrix according to preset convolution algorithm
Two eigenmatrixes;
Sorting module obtains third feature matrix for being ranked up recombination to the second characteristic matrix;
Extraction module obtains image for carrying out feature extraction to the third feature matrix according to the convolution algorithm
Characteristic set.
As an alternative embodiment, the conversion module includes:
Submodule is deviated, for carrying out data-bias processing to the fisrt feature matrix according to preset convolution algorithm,
Obtain side-play amount set;
Transform subblock, for generating offset moment matrix according to the side-play amount set;
Extracting sub-module is obtained for carrying out feature extraction to the fisrt feature matrix according to the offset moment matrix
Second characteristic matrix.
As an alternative embodiment, the sorting module is specifically used for carrying out matrix to the second characteristic matrix
Transposition processing, obtains third feature matrix.
The third aspect, the present invention provides a kind of computer equipment, the computer equipment includes memory and processing
Device, the memory run the computer program so that the computer is set for storing computer program, the processor
It is standby to execute the feature extracting method based on artificial intelligence image recognition described in first aspect present invention.
Fourth aspect, the present invention provides a kind of computer readable storage mediums, are stored with third aspect present invention institute
Computer program used in the computer equipment stated.
The feature extracting method and device based on artificial intelligence image recognition provided according to the present invention, can preferentially obtain
The fisrt feature matrix of images to be recognized, and first time convolutional calculation is carried out to the fisrt feature matrix, obtain second feature square
Battle array, so that second characteristic matrix, which can be used as intermediary matrix, provides benchmark for subsequent single calculation;It is obtained in second characteristic matrix
After getting, processing is ranked up to second characteristic matrix, third feature matrix corresponding with second characteristic matrix is obtained, makes
Obtaining third feature matrix may be directly applied to above-mentioned subsequent word calculating;Finally, again by convolution algorithm directly to the
Three eigenmatrixes carry out only needing primary feature extraction, to get characteristics of image set.As it can be seen that implementing this embodiment party
Formula can be avoided by performing corresponding processing to the fisrt feature matrix with identification image to every in the fisrt feature matrix
A element extracts the repetitive operation of calculating, and in the subsequent second characteristic matrix obtained according to above-mentioned processing and third feature
Matrix extracts to obtain characteristics of image set, so as to pass through the work for reducing feature extraction in artificial intelligence model establishment process
It measures, and then improves the efficiency of feature extraction and the efficiency of artificial intelligence model foundation.
To enable the above objects, features and advantages of the present invention 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
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of the scope of the invention.
Fig. 1 is a kind of stream for feature extracting method based on artificial intelligence image recognition that first embodiment of the invention provides
Journey schematic diagram;
Fig. 2 is a kind of stream for feature extracting method based on artificial intelligence image recognition that second embodiment of the invention provides
Journey schematic diagram;
Fig. 3 is a kind of knot for feature deriving means based on artificial intelligence image recognition that third embodiment of the invention provides
Structure schematic diagram;
Fig. 4 is a kind of velocity contrast applied to feature extraction in face recognition algorithms that second embodiment of the invention provides
Figure;
Fig. 5 is a kind of average time applied to feature extraction in face recognition algorithms that second embodiment of the invention provides
Comparison diagram.
Specific embodiment
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed invention to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
Range, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
According to the feature extracting method and device provided by the present application based on artificial intelligence image recognition, can preferentially obtain
The fisrt feature matrix of images to be recognized, and first time convolutional calculation is carried out to the fisrt feature matrix, obtain second feature square
Battle array, so that second characteristic matrix, which can be used as intermediary matrix, provides benchmark for subsequent single calculation;It is obtained in second characteristic matrix
After getting, processing is ranked up to second characteristic matrix, third feature matrix corresponding with second characteristic matrix is obtained, makes
Obtaining third feature matrix may be directly applied to above-mentioned subsequent word calculating;Finally, again by convolution algorithm directly to the
Three eigenmatrixes carry out only needing primary feature extraction, to get characteristics of image set.As it can be seen that implementing this embodiment party
Formula can be avoided by performing corresponding processing to the fisrt feature matrix with identification image to every in the fisrt feature matrix
A element extracts the repetitive operation of calculating, and in the subsequent second characteristic matrix obtained according to above-mentioned processing and third feature
Matrix extracts to obtain characteristics of image set, so as to pass through the work for reducing feature extraction in artificial intelligence model establishment process
It measures, and then improves the efficiency of feature extraction and the efficiency of artificial intelligence model foundation.It is retouched below by embodiment
It states.
Wherein, above-mentioned technical method can also be realized using relevant software or hardware, in this present embodiment
No longer add to repeat.For the feature extracting method and device based on artificial intelligence image recognition, below by embodiment into
Row description.
Embodiment 1
Referring to Fig. 1, being a kind of stream of feature extracting method based on artificial intelligence image recognition provided in this embodiment
Journey schematic diagram, should feature extracting method based on artificial intelligence image recognition the following steps are included:
S101, fisrt feature matrix is obtained, fisrt feature matrix includes the feature of images to be recognized.
In the present embodiment, images to be recognized is arbitrary image, to not being limited in any way in this present embodiment.
In the present embodiment, images to be recognized is defaulted as getting, for the acquisition methods of the images to be recognized
It is not limited in any way, wherein acquisition methods may include shooting acquisition methods, laser acquisition method etc..
In the present embodiment, fisrt feature matrix be can be by being sampled, being analyzed to images to be recognized or other hands
Section obtains, to not being limited in any way in this present embodiment.
In the present embodiment, fisrt feature matrix can be the color square that the color conversion that images to be recognized kind include obtains
Battle array, wherein include the element that can be largely converted to color in the matrix.
In the present embodiment, the matrix that fisrt feature matrix can also be formed for a part of feature of images to be recognized,
Middle specific features are the feature that certain element is converted in images to be recognized, wherein above-mentioned element can be above-mentioned color,
Or corresponding number mark of information importance level etc..
S102, feature extraction is carried out to fisrt feature matrix according to preset convolution algorithm, obtains second characteristic matrix.
As an alternative embodiment, feature extraction is carried out to fisrt feature matrix according to preset convolution algorithm,
The step of obtaining second characteristic matrix include:
Obtain the convolution kernel that preset convolution algorithm includes;
Process of convolution is carried out to fisrt feature matrix according to the convolution kernel and obtains multiple second feature, and is special by multiple second
Sign is combined into second characteristic matrix, wherein the second characteristic matrix is continuously arranged multiple second feature combinations.
Implement this embodiment, fisrt feature matrix can be converted according to convolution kernel as multiple second feature, and according to
Multiple second feature combine to obtain a second characteristic matrix.For example, which includes multirow, wherein every
A line is all the second feature obtained by a convolution nuclear convolution.
In the present embodiment, convolution algorithm includes convolution kernel, wherein convolution algorithm is to be rolled up according to the convolution kernel to matrix
The algorithm of product processing.
In the present embodiment, second characteristic matrix is the convolution results of fisrt feature matrix, is referred to as intermediate result, should
Convolution results are not subsequent characteristics of image set, and entirely different in the sense with subsequent characteristics of image set yet.
S103, recombination is ranked up to second characteristic matrix, obtains third feature matrix.
It may include calculating process during sequence recombination in the present embodiment.
As an alternative embodiment, being ranked up recombination to second characteristic matrix, third feature matrix is obtained
Step may include:
Calculating is optimized to second characteristic matrix, obtains optimization eigenmatrix, and carry out according to the optimization eigenmatrix
It is converted to third feature matrix.
Implement this embodiment, the workload of subsequent conversion can be reduced, to improve efficiency.
In the present embodiment, sequence recombination is to be directed to second characteristic matrix, therefore sequence recombination refers to second feature
The numeric sorting recombination method of matrix, that is to say, that the method for sequence recombination is the method that mathematics adjustment calculates.
In the present embodiment, the method for sequence recombination is preferably the method for T transposition.
In the present embodiment, third feature matrix is the matrix for calculating characteristics of image, this non-final result.
S104, feature extraction is carried out to third feature matrix according to convolution algorithm, obtains characteristics of image set.
In the present embodiment, feature combinations include multiple images feature, wherein each characteristics of image refers to figure to be identified
A feature or multiple features as in, for not making any limit in concrete condition the present embodiment of each characteristics of image reference
It is fixed.
As an alternative embodiment, carrying out feature extraction to third feature matrix according to convolution algorithm, figure is obtained
After as the step of characteristic set, this method further include:
The characteristics of image set is adjusted, image characteristic matrix is obtained;
It determines that the image characteristic matrix is fisrt feature matrix, and triggers and execute above-mentioned step S101.
Implement this embodiment, step S101 can be repeated, so that the available circulation of this feature extracting method,
To improve the precision of feature extraction.
In the present embodiment, the step of this method concentrates on the feature extraction of artificial intelligence image recognition, that is to say, that the party
Method is intended to the step of optimizing feature extraction, specifically, this method is the arithmetic speed of the convolutional layer in optimization artificial intelligence model,
It can be seen that this method can also reach the deduction of feature extraction efficiency by optimizing convolutional layer processing speed, to improve people
Work model of mind establishes efficiency (wherein principle is to reduce original repeated workload).
In the present embodiment, this method is preferably applied in artificial intelligence image recognition, and convolutional layer extracts feature
Process.
In the present embodiment, this method is preferably applied in the intelligent algorithm of recognition of face.
In the feature extracting method based on artificial intelligence image recognition described in Fig. 1, can preferentially it obtain to be identified
The fisrt feature matrix of image, and first time convolutional calculation is carried out to the fisrt feature matrix, second characteristic matrix is obtained, so that
Second characteristic matrix can be used as intermediary matrix and provide benchmark for subsequent single calculation;It is got in second characteristic matrix
Afterwards, processing is ranked up to second characteristic matrix, third feature matrix corresponding with second characteristic matrix is obtained, so that third
Eigenmatrix may be directly applied to above-mentioned subsequent word and calculate;Finally, again by convolution algorithm directly to third feature
Matrix carries out only needing primary feature extraction, to get characteristics of image set.As it can be seen that it is described based on people to implement Fig. 1
The feature extracting method of work intelligent image identification, can be by performing corresponding processing the fisrt feature matrix with identification image
It avoids extracting each element in the fisrt feature matrix repetitive operation of calculating, and is handled subsequent according to above-mentioned
To second characteristic matrix and third feature matrix extract to obtain characteristics of image set, so as to by reducing artificial intelligence
The workload of feature extraction during model foundation, so improve feature extraction efficiency and artificial intelligence model establish
Efficiency.
Embodiment 2
Referring to Fig. 2, Fig. 2 is a kind of feature extracting method based on artificial intelligence image recognition provided in this embodiment
Flow diagram.As shown in Fig. 2, should feature extracting method based on artificial intelligence image recognition the following steps are included:
S201, fisrt feature matrix is obtained, fisrt feature matrix includes the feature of images to be recognized.
In the present embodiment, identical step or operation can be used in conjunction with identical explanation in different embodiments, right
This, no longer adds to repeat in the present embodiment.
S202, data-bias processing is carried out to fisrt feature matrix according to preset convolution algorithm, obtains side-play amount set.
In the present embodiment, migration processing is carried out to fisrt feature matrix according to the convolution kernel that convolution algorithm includes, wherein should
Migration processing refers to the processing method of the above-mentioned second feature of convolution kernel offset extraction in process of convolution.
In the present embodiment, multiple offsets that arrive of migration processing, and group are carried out to fisrt feature matrix by convolution kernel
It closes multiple offset and obtains side-play amount set.Wherein, which includes multiple offsets, and each offset all can be with
It is interpreted as deviant.
S203, offset moment matrix is generated according to side-play amount set.
As an alternative embodiment, before the step of generating offset moment matrix according to side-play amount set, the side
Method further include:
Pre-establish a vacant matrix.
Implement this embodiment, storage location can be provided for subsequent offset moment matrix, to meet computer fortune
Row condition or the occupancy for reducing memory.
In the present embodiment, for saving each second feature, (second characteristic matrix each of includes above-mentioned control matrix
Element) offset in fisrt feature matrix.
For example, which can be understood as referring to establish above-mentioned vacant matrix after, algorithm is initialized,
Operation primary above-mentioned the step of offset is put into above-mentioned control matrix.
S204, feature extraction is carried out to fisrt feature matrix according to offset moment matrix, obtains second characteristic matrix.
Implement this embodiment, it can be (vacant here to original vacant second characteristic matrix by offset moment matrix
Second characteristic matrix refer to do not generate second characteristic matrix before preset one empty matrix ghost) carry out quick valuation,
To obtain second characteristic matrix.
S205, the processing of matrix transposition is carried out to second characteristic matrix, obtains third feature matrix.
In the present embodiment, transposition processing is well known matrix transposition method, to no longer adding to repeat in this present embodiment.
In the present embodiment, third feature matrix and the phase of second characteristic matrix are almost only in that " transposition ".
S206, the convolution kernel that convolution algorithm includes is obtained.
In the present embodiment, convolution algorithm includes its specific convolution kernel, no longer adds to repeat in the present embodiment.
S207, matrix multiple operation is carried out to convolution kernel and third feature matrix, obtains characteristics of image set.
Implement this embodiment, matrix multiplication can be done by convolution kernel and third feature matrix in operation
Realization successively obtains respective element from third feature matrix, disappears to reduce jump offset bring operation in conventional method
Consumption, to improve efficiency.
In the present embodiment, characteristics of image set refers to the result that convolution obtains, it is to be understood that the characteristics of image set
It is the set of the feature of images to be recognized.
S208, judge whether characteristics of image set meets default feature request, if so, terminating this process;If it is not, then holding
Row step S209~S210.
Implement this embodiment, the process of above-mentioned convolution can be superimposed, realize the superposition of multiple convolutional layers, wherein presetting
Feature request is whether judging characteristic needs further convolution.
In the present embodiment, default feature request can be greater than preset quantity with the requirement of quantity, such as characteristic quantity;Default feature
It is required that can also be the requirement of result fine degree, as feature will meet whether preset fine degree (can be understood as convolution
In place).
S209, pond processing is carried out to characteristics of image set, obtains pond characteristic set.
In the present embodiment, pondization processing can be by pond layer on behalf of progress.
In the present embodiment, pondization processing can be maximum pond.
S210, pond characteristic set is determined as fisrt feature matrix, and triggers and executes step S202.
Implement this embodiment, execution can be superimposed, improves the order of accuarcy of identification.
Implement this embodiment, can be superimposed by the convolutional layer and pond layer of feature extraction, realize one layer of convolution
The feature extraction mode in one layer of pond, to realize the effect of above-mentioned raising order of accuarcy.
Referring to Fig. 4, when being applied to as seen from Figure 4 this Figure illustrates this method in the intelligent algorithm of recognition of face,
Velocity contrast's situation of feature extraction.Wherein, the curve being located above in the figure is to carry out spy using traditional CNN convolution algorithm
Sign required rate curve when extracting, and underlying curve is required when carrying out feature extraction in this method in figure
Rate curve;As it can be seen that the feature extracting method that the present embodiment is proposed can get desired feature faster, to save
Time improves efficiency, and then realizes that artificial intelligence model establishes the shortening of required time.
Referring to Fig. 5, the figure describes this method and applies in the intelligent algorithm of recognition of face as seen from Figure 5
When, the average time comparison diagram of feature extraction.Wherein, left is that method described in this implementation is not used to use conventional method
The feature extraction mean consumption time, and right be the feature extraction mean consumption using method described in the present embodiment when
Between;As it can be seen that the effect being intended by with Fig. 4 can be played using method as described in this embodiment, to improve relevant
Calculating and service efficiency.
As it can be seen that each element of second characteristic matrix can be calculated with respect to the first spy by pretreatment in advance by implementing this method
Levy the deviation post of matrix;And in operation, by above-mentioned deviation post, directly to second characteristic matrix assignment;Herein it
Afterwards, pretreatment operation is carried out herein, so that second characteristic matrix transposition, thus the fortune for value of jumping when further element being avoided to be multiplied
Consumption is calculated, therefore this method compares traditional CNN convolution algorithm, reduces operation times and memory offset addressing operation, promoted
Operation efficiency.
As it can be seen that implement the feature extracting method based on artificial intelligence image recognition described in Fig. 2, it can be by knowing to band
The fisrt feature matrix of other image performs corresponding processing to avoid extracting meter to each element in the fisrt feature matrix
The repetitive operation of calculation, and extract to obtain figure in the subsequent second characteristic matrix obtained according to above-mentioned processing and third feature matrix
As characteristic set, so as to pass through the workload for reducing feature extraction in artificial intelligence model establishment process, and then improve
The efficiency that the efficiency and artificial intelligence model of feature extraction are established.
Embodiment 3
Referring to Fig. 3, being a kind of knot of feature deriving means based on artificial intelligence image recognition provided in this embodiment
Structure schematic diagram.
As shown in figure 3, the feature deriving means based on artificial intelligence image recognition include:
Module 310 is obtained, for obtaining fisrt feature matrix, the fisrt feature matrix includes the spy of images to be recognized
Sign;
Conversion module 320 obtains second for carrying out feature extraction to fisrt feature matrix according to preset convolution algorithm
Eigenmatrix;
Sorting module 330 obtains third feature matrix for being ranked up recombination to second characteristic matrix;
Extraction module 340 obtains set of image characteristics for carrying out feature extraction to third feature matrix according to convolution algorithm
It closes.
As an alternative embodiment, conversion module 320 includes:
Submodule 321 is deviated, for carrying out data-bias processing to fisrt feature matrix according to preset convolution algorithm, is obtained
To side-play amount set;
Transform subblock 322, for generating offset moment matrix according to side-play amount set;
Extracting sub-module 323 obtains the second spy for carrying out feature extraction to fisrt feature matrix according to offset moment matrix
Levy matrix.
As an alternative embodiment, sorting module 330 is specifically used for carrying out matrix transposition to second characteristic matrix
Processing, obtains third feature matrix.
In the present embodiment, being somebody's turn to do the feature deriving means based on artificial intelligence image recognition can be used institute in above-described embodiment
The feature extracting method based on artificial intelligence image recognition of description, and explanation described in above-described embodiment is continued to use,
Therefore in the present embodiment, will not be described in great detail for illustrating accordingly.
As it can be seen that implementing the feature deriving means based on artificial intelligence image recognition described in the present embodiment, can pass through
Fisrt feature matrix with identification image is performed corresponding processing to avoid carrying out element each in the fisrt feature matrix
The repetitive operation calculated is extracted, and is extracted in the subsequent second characteristic matrix obtained according to above-mentioned processing and third feature matrix
Characteristics of image set is obtained, to reduce the workload of feature extraction, and then reduces the work of artificial intelligence model foundation
Amount, the efficiency that the efficiency and artificial intelligence model for improving feature extraction are established.It is described below by embodiment.
In addition, the computer equipment may include smart phone, put down the present invention also provides another computer equipment
Plate computer, vehicle-mounted computer, intelligent wearable device etc..The computer equipment includes memory and processor, and memory can be used for depositing
Store up computer program, processor by running above-mentioned computer program, thus make computer equipment execute the above method or on
State the function of each unit in device.
Memory may include storing program area and storage data area, wherein storing program area can storage program area, at least
Application program needed for one function (such as sound-playing function, image player function etc.) etc.;Storage data area can store root
Created data (such as audio data, phone directory etc.) etc. are used according to computer equipment.In addition, memory may include height
Fast random access memory, can also include nonvolatile memory, a for example, at least disk memory, flush memory device,
Or other volatile solid-state parts.
The present embodiment additionally provides a kind of computer storage medium, for storing calculating used in above-mentioned computer equipment
Machine program.
In several embodiments provided herein, it should be understood that disclosed device and method can also pass through
Other modes are realized.The apparatus embodiments described above are merely exemplary, for example, flow chart and structure in attached drawing
Figure shows the system frame in the cards of the device of multiple embodiments according to the present invention, method and computer program product
Structure, function and operation.In this regard, each box in flowchart or block diagram can represent a module, section or code
A part, a part of above-mentioned module, section or code includes one or more for implementing the specified logical function
Executable instruction.It should also be noted that function marked in the box can also be to be different from the implementation as replacement
The sequence marked in attached drawing occurs.For example, two continuous boxes can actually be basically executed in parallel, they are sometimes
It can execute in the opposite order, this depends on the function involved.It is also noted that in structure chart and/or flow chart
The combination of each box and the box in structure chart and/or flow chart, can function or movement as defined in executing it is dedicated
Hardware based system realize, or can realize using a combination of dedicated hardware and computer instructions.
In addition, each functional module or unit in each embodiment of the present invention can integrate one independence of formation together
Part, be also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
If described function is realized and when sold or used as an independent product in the form of software function module, can
To be stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention substantially or
Say that the part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, it should
Computer software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be
Smart phone, personal computer, server or network equipment etc.) execute the complete of method described by each embodiment of the present invention
Portion or part steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.
Content described above, only a specific embodiment of the invention, but protection scope of the present invention is not limited to
In this, anyone skilled in the art in the technical scope disclosed by the present invention, can readily occur in variation or replace
It changes, should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (10)
1. a kind of feature extracting method based on artificial intelligence image recognition characterized by comprising
Fisrt feature matrix is obtained, the fisrt feature matrix includes the feature of images to be recognized;
Feature extraction is carried out to the fisrt feature matrix according to preset convolution algorithm, obtains second characteristic matrix;
Recombination is ranked up to the second characteristic matrix, obtains third feature matrix;
Feature extraction is carried out to the third feature matrix according to the convolution algorithm, obtains characteristics of image set.
2. the feature extracting method according to claim 1 based on artificial intelligence image recognition, which is characterized in that described
Feature extraction is carried out to the fisrt feature matrix according to preset convolution algorithm, the step of obtaining second characteristic matrix includes:
Data-bias processing is carried out to the fisrt feature matrix according to preset convolution algorithm, obtains side-play amount set;
Offset moment matrix is generated according to the side-play amount set;
Feature extraction processing is carried out to the fisrt feature matrix according to the offset moment matrix, obtains second characteristic matrix.
3. the feature extracting method according to claim 1 based on artificial intelligence image recognition, which is characterized in that described right
The step of second characteristic matrix is ranked up recombination, obtains third feature matrix include:
The processing of matrix transposition is carried out to the second characteristic matrix, obtains third feature matrix.
4. the feature extracting method according to claim 1 based on artificial intelligence image recognition, which is characterized in that described
Feature extraction is carried out to the third feature matrix according to the convolution algorithm, the step of obtaining characteristics of image set includes:
Obtain the convolution kernel that the convolution algorithm includes;
Matrix multiple operation is carried out to the convolution kernel and the third feature matrix, obtains characteristics of image set.
5. the feature extracting method according to claim 1 based on artificial intelligence image recognition, which is characterized in that the side
Method further include:
Judge whether described image characteristic set meets default feature request;
When described image characteristic set does not meet default feature request, pond processing is carried out to described image characteristic set, is obtained
To pond characteristic set;
The pond characteristic set is determined as the fisrt feature matrix, and execute it is described according to preset convolution algorithm to institute
State the step of fisrt feature matrix carries out feature extraction, obtains second characteristic matrix.
6. a kind of feature deriving means based on artificial intelligence image recognition characterized by comprising
Module is obtained, for obtaining fisrt feature matrix, the fisrt feature matrix includes the feature of images to be recognized;
Conversion module obtains the second spy for carrying out feature extraction to the fisrt feature matrix according to preset convolution algorithm
Levy matrix;
Sorting module obtains third feature matrix for being ranked up recombination to the second characteristic matrix;
Extraction module obtains characteristics of image for carrying out feature extraction to the third feature matrix according to the convolution algorithm
Set.
7. the feature deriving means according to claim 6 based on artificial intelligence image recognition, which is characterized in that modulus of conversion
Block includes:
Submodule is deviated, for carrying out data-bias processing to the fisrt feature matrix according to preset convolution algorithm, is obtained
Side-play amount set;
Transform subblock, for generating offset moment matrix according to the side-play amount set;
Extracting sub-module obtains second for carrying out feature extraction to the fisrt feature matrix according to the offset moment matrix
Eigenmatrix.
8. the feature deriving means according to claim 6 based on artificial intelligence image recognition, which is characterized in that the row
Sequence module is specifically used for carrying out the processing of matrix transposition to the second characteristic matrix, obtains third feature matrix.
9. a kind of computer equipment, which is characterized in that including memory and processor, the memory is for storing computer
Program, the processor runs the computer program so that the computer equipment executes according to claim 1 to any in 5
The feature extracting method based on artificial intelligence image recognition described in.
10. a kind of computer readable storage medium, which is characterized in that it is stored in computer equipment as claimed in claim 9
Used computer program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910162758.8A CN109815969A (en) | 2019-03-05 | 2019-03-05 | A kind of feature extracting method and device based on artificial intelligence image recognition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910162758.8A CN109815969A (en) | 2019-03-05 | 2019-03-05 | A kind of feature extracting method and device based on artificial intelligence image recognition |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109815969A true CN109815969A (en) | 2019-05-28 |
Family
ID=66608102
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910162758.8A Pending CN109815969A (en) | 2019-03-05 | 2019-03-05 | A kind of feature extracting method and device based on artificial intelligence image recognition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109815969A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112233353A (en) * | 2020-09-24 | 2021-01-15 | 国网浙江兰溪市供电有限公司 | Artificial intelligence-based anti-fishing monitoring system and monitoring method thereof |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105224963A (en) * | 2014-06-04 | 2016-01-06 | 华为技术有限公司 | The method of changeable degree of depth learning network structure and terminal |
CN106055576A (en) * | 2016-05-20 | 2016-10-26 | 大连理工大学 | Rapid and effective image retrieval method under large-scale data background |
CN106504266A (en) * | 2016-09-29 | 2017-03-15 | 北京市商汤科技开发有限公司 | The Forecasting Methodology of walking behavior and device, data processing equipment and electronic equipment |
CN106898011A (en) * | 2017-01-06 | 2017-06-27 | 广东工业大学 | A kind of method that convolutional neural networks convolution nuclear volume is determined based on rim detection |
CN107092870A (en) * | 2017-04-05 | 2017-08-25 | 武汉大学 | A kind of high resolution image semantics information extracting method and system |
CN107220706A (en) * | 2016-12-29 | 2017-09-29 | 恩泊泰(天津)科技有限公司 | Vehicle-mounted deep neural network optimization method based on compression of parameters and structure compresses |
CN108122030A (en) * | 2016-11-30 | 2018-06-05 | 华为技术有限公司 | A kind of operation method of convolutional neural networks, device and server |
CN108520204A (en) * | 2018-03-16 | 2018-09-11 | 西北大学 | A kind of face identification method |
CN108919229A (en) * | 2018-10-09 | 2018-11-30 | 中国科学院沈阳自动化研究所 | A kind of matrix reconstruction imaging method based on convolution inverse projection |
CN109325589A (en) * | 2017-07-31 | 2019-02-12 | 华为技术有限公司 | Convolutional calculation method and device |
CN109376695A (en) * | 2018-11-26 | 2019-02-22 | 北京工业大学 | A kind of smog detection method based on depth hybrid neural networks |
-
2019
- 2019-03-05 CN CN201910162758.8A patent/CN109815969A/en active Pending
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105224963A (en) * | 2014-06-04 | 2016-01-06 | 华为技术有限公司 | The method of changeable degree of depth learning network structure and terminal |
CN106055576A (en) * | 2016-05-20 | 2016-10-26 | 大连理工大学 | Rapid and effective image retrieval method under large-scale data background |
CN106504266A (en) * | 2016-09-29 | 2017-03-15 | 北京市商汤科技开发有限公司 | The Forecasting Methodology of walking behavior and device, data processing equipment and electronic equipment |
CN108122030A (en) * | 2016-11-30 | 2018-06-05 | 华为技术有限公司 | A kind of operation method of convolutional neural networks, device and server |
CN107220706A (en) * | 2016-12-29 | 2017-09-29 | 恩泊泰(天津)科技有限公司 | Vehicle-mounted deep neural network optimization method based on compression of parameters and structure compresses |
CN106898011A (en) * | 2017-01-06 | 2017-06-27 | 广东工业大学 | A kind of method that convolutional neural networks convolution nuclear volume is determined based on rim detection |
CN107092870A (en) * | 2017-04-05 | 2017-08-25 | 武汉大学 | A kind of high resolution image semantics information extracting method and system |
CN109325589A (en) * | 2017-07-31 | 2019-02-12 | 华为技术有限公司 | Convolutional calculation method and device |
CN108520204A (en) * | 2018-03-16 | 2018-09-11 | 西北大学 | A kind of face identification method |
CN108919229A (en) * | 2018-10-09 | 2018-11-30 | 中国科学院沈阳自动化研究所 | A kind of matrix reconstruction imaging method based on convolution inverse projection |
CN109376695A (en) * | 2018-11-26 | 2019-02-22 | 北京工业大学 | A kind of smog detection method based on depth hybrid neural networks |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112233353A (en) * | 2020-09-24 | 2021-01-15 | 国网浙江兰溪市供电有限公司 | Artificial intelligence-based anti-fishing monitoring system and monitoring method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108416440A (en) | A kind of training method of neural network, object identification method and device | |
CN108573246B (en) | Time sequence action identification method based on deep learning | |
CN109241903A (en) | Sample data cleaning method, device, computer equipment and storage medium | |
CN110991362A (en) | Pedestrian detection model based on attention mechanism | |
WO2022198636A1 (en) | Memory allocation method for ai processor, computer device, and computer-readable storage medium | |
CN110378348A (en) | Instance of video dividing method, equipment and computer readable storage medium | |
CN111400535A (en) | Lightweight face recognition method, system, computer device and storage medium | |
CN110222598A (en) | A kind of video behavior recognition methods, device, storage medium and server | |
CN110210278A (en) | A kind of video object detection method, device and storage medium | |
CN111612024B (en) | Feature extraction method, device, electronic equipment and computer readable storage medium | |
CN111127309B (en) | Portrait style migration model training method, portrait style migration method and device | |
CN107292458A (en) | A kind of Forecasting Methodology and prediction meanss applied to neural network chip | |
CN106227792A (en) | Method and apparatus for pushed information | |
CN106778910A (en) | Deep learning system and method based on local training | |
CN111695502A (en) | Feature updating method and device for face recognition and computer equipment | |
CN110321892A (en) | A kind of picture screening technique, device and electronic equipment | |
CN109829371A (en) | A kind of method for detecting human face and device | |
CN114881225A (en) | Power transmission and transformation inspection model network structure searching method, system and storage medium | |
Al-Amaren et al. | RHN: A residual holistic neural network for edge detection | |
CN114783021A (en) | Intelligent detection method, device, equipment and medium for wearing of mask | |
CN109165654A (en) | The training method and object localization method and device of a kind of target location model | |
CN109815969A (en) | A kind of feature extracting method and device based on artificial intelligence image recognition | |
CN111709415A (en) | Target detection method, target detection device, computer equipment and storage medium | |
CN117193992B (en) | Model training method, task scheduling device and computer storage medium | |
CN113361567A (en) | Image processing method, image processing device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190528 |