TOP > クイック検索 > 外国特許検索 > CHARACTERISTIC AMOUNT CONVERSION MODULE, PATTERN IDENTIFICATION DEVICE, PATTERN IDENTIFICATION METHOD, AND PROGRAM

CHARACTERISTIC AMOUNT CONVERSION MODULE, PATTERN IDENTIFICATION DEVICE, PATTERN IDENTIFICATION METHOD, AND PROGRAM NEW

外国特許コード F170009161
整理番号 (S2016-0420-N0)
掲載日 2017年9月6日
出願国 世界知的所有権機関(WIPO)
国際出願番号 2017JP005600
国際公開番号 WO 2017141997
国際出願日 平成29年2月15日(2017.2.15)
国際公開日 平成29年8月24日(2017.8.24)
優先権データ
  • 特願2016-025892 (2016.2.15) JP
発明の名称 (英語) CHARACTERISTIC AMOUNT CONVERSION MODULE, PATTERN IDENTIFICATION DEVICE, PATTERN IDENTIFICATION METHOD, AND PROGRAM NEW
発明の概要(英語) A characteristic amount conversion module (characteristic amount conversion means) (500) to which the present invention is applied is equipped with: a learning data input unit (510); an intermediate layer setting unit (520), which inputs learning data that has been input to the learning data input unit into a learning model configured from a function comprising a formula having, as factors, a weighting and a bias having a second dimensional number satisfying an optical calculation constraint condition, and which sets an intermediate layer having the second dimensional number; an output calculation unit (530); an error calculation unit (540); a determination unit (550); a weighting bias changing unit (560); a learning model output unit (570); and a control unit (580).
特許請求の範囲(英語) [claim1]
1. Being the feature quantitative conversion module which converts the feature quantity of the study data with the study in neural network,
The study data entry section where the number of optional first dimensions and the study data which possesses the optional feature quantity are input and,
The center setting section which sets the center which is formed inputs the aforementioned study data which is input into the aforementioned study data entry section into the study model which, from the function which consists of the formula which has with weight and the bias which possess the number of back burner origins which satisfy specified optical operational constraint in factor possesses the number of aforementioned back burner origins and,
The output calculation section which calculates the output layer which possesses the number of aforementioned first dimensions from the aforementioned center which is set with the aforementioned center setting section calculates the output of the aforementioned neural network and,
The error calculation section which calculates the error of output and the aforementioned study data of the aforementioned neural network which was calculated by the aforementioned output calculation section and,
The decision section which is decided whether or not the aforementioned error which was calculated by the aforementioned error calculation section satisfies specified condition, and,
Weight bias modification section which modifies with the aforementioned weight and the aforementioned bias respectively on the basis of the aforementioned error, the aforementioned weight after the modifying and obtains the aforementioned bias and,
The study model output section which outputs the study model which is formed by the function which consists of the formula which has with the aforementioned weight and the aforementioned bias after the description above modifying in factor and,
When it is decided, that the aforementioned error satisfies specified condition, in the aforementioned decision section the study model which is formed by the function which consists of the formula which has with the aforementioned weight and the aforementioned bias after the description above modifying in factor from the aforementioned output section is made to output, when it is decided, that the aforementioned error does not satisfy specified condition, in the aforementioned decision section the aforementioned weight which is included in the aforementioned system of the aforementioned study model in the aforementioned output calculation section and the description above which can obtain the aforementioned bias in the aforementioned weight bias modification section it modifies respectively in the aforementioned weight and the aforementioned bias after the modifying, calculating the output of the aforementioned neural networkThe control section which it can point and,
Having,
As for the aforementioned optical operational constraint, the frequency of specified quantization in optical calculating which is done making use of the output of the aforementioned neural network, the feature quantitative conversion module which features that it is decided by the optical parameter of the optical correlator which is used for the aforementioned optical operation, and the number of specified energies.
[claim2]
2. As for the aforementioned optical correlator, the indicating element which indicates the data for optical operation which is based on the output of the aforementioned neural network and the light where the data for the aforementioned operation is reflected the lens which condenses and, having,
In the aforementioned optical parameter, in the claim 1 which features the pixel pitch of the aforementioned indicating element and that radius of the aforementioned lens is included feature quantitative conversion module of statement.
[claim3]
3. The aforementioned optical correlator reduces the beam diameter of the light where the data for the aforementioned optical operation which is indicated in the aforementioned indicating element is reflected, directing the light where the said beam diameter is reduced to the aforementioned lens, radiation the reduction optical system which is done furthermore having,
In the aforementioned optical parameter, in the claim 2 which features that reduction magnification ratio of the aforementioned reduction optical system is included feature quantitative conversion module of statement.
[claim4]
4. As for the number of aforementioned back burner origins, in the aforementioned claim 3 which features that (1) formula below is satisfied feature quantitative conversion module of statement.
Furthermore, as for z the number of aforementioned back burner origins, as for a pixel pitch of the aforementioned indicating element [mm], as for r radius of the aforementioned lens [mm], as for b reduction magnification ratio of the aforementioned reduction optical system is shown in the above-mentioned (1) formula.
[claim5]
5. From claim 1 in either of claim 4 feature quantitative conversion module of statement and,
The study expedient which studies the feature kind of ejector which converts the identification precision of the data for the aforementioned study which designates the data for study where the class label is added as the entrance mosquito, has the constraint for optical operation maximally and,
From the aforementioned feature ejector the data converter for optical operation which converts the feature quantity which is extracted to the feature ejector and,
The optical memory section which remembers the data for optical operation and the optical operational expedient due to the similarity calculation section which offers optical correlation function similarity in the data hearing for plural optical operations by calculating and,
It converts to the one for optical operation, the register expedient which is registered to the optical memory section where it inputs the data for register, which becomes the dictionary of the identification object data the feature quantity which is extracted making use of the feature ejector which is obtained by the aforementioned study expedient, in the data converter for the aforementioned optical operation in the aforementioned optical operational expedient as the feature quantity for optical operation and,
The identification expedient where it inputs the identification object data, it calculates in the similarity calculation section in similarity of the feature quantitative hearing which is registered to the optical memory section in the feature quantity and the aforementioned register expedient which are extracted from the feature ejector which is formed with the aforementioned study expedient the aforementioned optical operational expedient from the output section outputs the identification result and,
The pattern recognition device which features that it has.
[claim6]
6. The study expedient which studies the feature kind of ejector which converts the identification precision of the data for the aforementioned study which designates the data for study where the class label is added as the entrance mosquito, has the constraint for optical operation maximally and,
From the aforementioned feature ejector the data converter for optical operation and the optical memory section which remembers the data for optical operation and the optical operational expedient due to the similarity calculation section which offers optical correlation function in the data hearing for plural optical operations similarity the feature quantity which is extracted is converted to the feature ejector by calculating and,
It converts to the one for optical operation, the register expedient which is registered to the optical memory section where it inputs the data for register, which becomes the dictionary of the identification object data the feature quantity which is extracted making use of the feature ejector which is obtained by the aforementioned study expedient, in the data converter for the aforementioned optical operation in the aforementioned optical operational expedient as the feature quantity for optical operation and,
The identification expedient where it inputs the identification object data, it calculates in the similarity calculation section in similarity of the feature quantitative hearing which is registered to the optical memory section in the feature quantity and the aforementioned register expedient which are extracted from the feature ejector which is formed with the aforementioned study expedient the aforementioned optical operational expedient from the output section outputs the identification result and,
The pattern recognition device which features that it has.
[claim7]
7. In the aforementioned study expedient, in order the output by the feature ejector which is formed after the studying in the process of the study in forming the feature ejector, binary to reach the feature quantity which closely resembles, value of each dimension of the feature quantity like the step function which closely resembles to the activated function in aforementioned feature ejector formation department with step function, or sigmoid function the claim 5 which features that the function which it makes the range of value of 1 be distributed from 0 is applied or in claim 6 the pattern recognition device of statement.
[claim8]
8. In aforementioned study expedient, in order to reach the feature quantity where the output by the feature ejector which is formed after the studying in the process of the study in forming the feature ejector, binary reaches the feature quantity which closely resembles and by comparison with the case where it is many-valued the expression power which the tsu and the feature quantity have like float value even is not inferior, the number of parameter parts where renewal is done by the study of aforementioned feature ejector formation department, by comparison with the case where the feature ejector which extracts the many-valued feature quantity is formed, from the claim 5 which features that it makes big number claim 7In each case in 1 sections pattern recognition device of statement.
[claim9]
9. The result of study of the feature ejector formation department in the aforementioned study expedient the binary which from 0 where it is obtained from the feature ejector which is obtained the feature quantity of each dimension is distributed to the range of 1 the feature quantity which closely resembles, in the data converter for the aforementioned optical operation at black-market price of specification to binary from the claim 5 which features that it converts either of claim 8 in 1 sections the pattern recognition device of statement.
[claim10]
10. In the optical operational restriction addition section of the feature ejector formation department in the aforementioned study expedient, from the aforementioned study expedient the feature quantity which gives the input data vis-a-vis the feature quantitative ejector which is obtained, extracts the claim 5 which features that the technique which the weight vector which did the optimization by the linear discrimination circuit to binary is converted is had or in claim 6 the pattern recognition device of statement.
[claim11]
11. In the optical operational restriction addition section of the feature ejector formation department in the aforementioned study expedient, from the aforementioned study expedient the feature quantity which gives the input data vis-a-vis the feature quantitative ejector which is obtained, extracts, in plural party values, making use of the respective black-market price and small and large relationship of feature quantitative value of each dimension to binary in the claim 10 which features that the technique which is converted is had the pattern recognition device of statement.
[claim12]
12. As for the aforementioned optical memory section from the claim 5 which features that it is holographic record medium either of claim 11 in 1 sections the pattern recognition device of statement.
[claim13]
13. While the horogurabuitsuku record medium in the aforementioned optical memory section being formed by disk condition, turning the aforementioned holographic record medium, the aforementioned identification object data from the claim 5 which features that it irradiates either of claim 12 in 1 sections the pattern recognition device of statement.
[claim14]
14. In the aforementioned optical operational expedient, from the claim 5 which features that feature extraction the data converter for optical operation, with optical operation either of claim 13 in 1 sections the pattern recognition device of statement.
[claim15]
15. The study step which studies the feature kind of ejector which converts the identification precision of the data for the aforementioned study which inputs the data for study, where the class label by the aforementioned pattern recognition device in the pattern recognition method of the pattern recognition device which identifies the data which is input in a some class, is added has the constraint for optical operation maximally and,
From the aforementioned feature ejector the data converter for optical operation and the optical memory section which remembers the data for optical operation and to convert to the one for optical operation to input the data for register, which becomes the dictionary of optical operational step and the identification object data by the similarity calculation section which offers optical correlation function in the data hearing for plural optical operations similarity the feature quantity which is extracted is converted to the feature ejector by calculating the feature quantity which is extracted making use of the feature ejector which is obtained by the aforementioned study step, in the data converter for the aforementioned optical operation, as the feature quantity for optical operation the description aboveThe register step which is registered to the optical memory section in optical operational step and,
The identification step where it inputs the identification object data, it calculates in the similarity calculation section in similarity of the feature quantitative question which is registered to the optical memory section in the feature quantity and the aforementioned register step which are extracted from the feature ejector which is formed with the aforementioned study step the aforementioned optical operational step from the output section outputs the recognition result and,
The pattern recognition method of including.
[claim16]
16. The study step which studies the feature kind of ejector which converts the identification precision of the data for the aforementioned study which the computer, inputs the data for study, where the class label is added has the constraint for optical operation maximally and,
From the aforementioned feature ejector the data converter for optical operation and the optical memory section which remembers the data for optical operation and the optical operational step by the similarity calculation section which offers optical correlation function in the data hearing for plural optical operations similarity the feature quantity which is extracted is converted to the feature ejector by calculating and,
It converts to the one for optical operation, the register step which is registered to the optical memory section where it inputs the data for register, which becomes the dictionary of the identification object data the feature quantity which is extracted making use of the feature ejector which is obtained by the aforementioned study step, in the data converter for the aforementioned optical operation in the aforementioned optical operational step as the feature quantity for optical operation and,
The program where it inputs the identification object data, it calculates in the similarity calculation section in similarity of the feature quantitative hearing which is registered to the optical memory section in the feature quantity and the aforementioned register step which are extracted from the feature ejector which is formed with the aforementioned study step the aforementioned optical operational step comes out and as the identification step which outputs the identification result from the mosquito section functioning it can point.
  • 出願人(英語)
  • ※2012年7月以前掲載分については米国以外のすべての指定国
  • THE UNIVERSITY OF ELECTRO-COMMUNICATIONS
  • 発明者(英語)
  • WATANABE ERIKO
  • FUJIYOSHI HIRONOBU
  • TANIGUCHI YASUFUMI
  • IKEDA KANAMI
  • WAKITA SUGURU
  • SUZUKI HIDENORI
国際特許分類(IPC)
指定国 (WO2017141997)
National States: AE AG AL AM AO AT AU AZ BA BB BG BH BN BR BW BY BZ CA CH CL CN CO CR CU CZ DE DJ DK DM DO DZ EC EE EG ES FI GB GD GE GH GM GT HN HR HU ID IL IN IR IS JP KE KG KH KN KP KR KW KZ LA LC LK LR LS LU LY MA MD ME MG MK MN MW MX MY MZ NA NG NI NO NZ OM PA PE PG PH PL PT QA RO RS RU RW SA SC SD SE SG SK SL SM ST SV SY TH TJ TM TN TR TT TZ UA UG US UZ VC VN ZA ZM ZW
ARIPO: BW GH GM KE LR LS MW MZ NA RW SD SL SZ TZ UG ZM ZW
EAPO: AM AZ BY KG KZ RU TJ TM
EPO: AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
OAPI: BF BJ CF CG CI CM GA GN GQ GW KM ML MR NE SN ST TD TG
ライセンスをご希望の方、特許の内容に興味を持たれた方は、下記までご連絡ください

PAGE TOP

close
close
close
close
close
close