US 12,141,176 B2
Classification system and classification method
Jeffry Fernando, Osaka (JP); Hisaji Murata, Osaka (JP); Hideto Motomura, Kyoto (JP); and Yuya Sugasawa, Osaka (JP)
Assigned to PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO., LTD., Osaka (JP)
Appl. No. 17/762,886
Filed by Panasonic Intellectual Property Management Co., Ltd., Osaka (JP)
PCT Filed Aug. 28, 2020, PCT No. PCT/JP2020/032539
§ 371(c)(1), (2) Date Mar. 23, 2022,
PCT Pub. No. WO2021/070505, PCT Pub. Date Apr. 15, 2021.
Claims priority of application No. 2019-184815 (JP), filed on Oct. 7, 2019.
Prior Publication US 2022/0342913 A1, Oct. 27, 2022
Int. Cl. G06F 16/28 (2019.01)
CPC G06F 16/287 (2019.01) 16 Claims
OG exemplary drawing
 
1. A classification system comprising:
an input interface configured to receive an input of target data;
a classification circuit that includes a plurality of classifiers and an NOR circuit connecting in parallel to each of the plurality of classifiers, outputs from the plurality of classifiers being input into the NOR circuit, the classification circuit comprising a learned neural network configured to classify the target data into a class that includes any one of a plurality of classes, the plurality of classes including a plurality of corresponding classes each corresponding to a respective one of the plurality of classifiers included in the classification circuit, and another class not corresponding to any of the plurality of classifiers; and
one or more processors configured to calculate a feature amount of the target data, and determine whether it is possible that the target data is classified into a new class different from the plurality of classes based on a classification result by the classification circuit and the feature amount of the target data calculated by the one or more processors, wherein:
the classification circuit comprises a learned neural network, and
the classification circuit is configured to classify, by using the learned neural network, the target data into the new class when the feature amount of the target data differs from any feature amount of the target data that are classified into the class that includes any one of the plurality of classes.