Information processing device, determination method, and information processing program
A final determination result is derived in accordance with target data, in consideration of determination results given by determining sections. An information processing device includes: a reliability determining section that determines, in accordance with an inspection image, reliabilities of determination results given by determining sections each configured to determine a given determination matter in accordance with the inspection image; and a comprehensive determination section configured to determine the given determination matter with use of the determination results and the reliabilities.
1 . An information processing device comprising a processor which carries out:
a reliability determining step of determining, using single target data, reliabilities which are indices indicating degrees of certainties of determination results obtained, from the single target data, by determining a given determination matter by a plurality of determination methods that are different from each other; and
a comprehensive determination step of determining the given determination matter with use of the determination results and the reliabilities determined in the reliability determining step,
wherein, in the reliability determining step, the processor determines the plurality of reliabilities, in accordance with output values obtained by inputting the target data into a plurality of reliability prediction models respectively corresponding to the plurality of determination methods; and
wherein each of the plurality of reliability prediction models is constructed by machine learning that uses training data in which (i) data used for determination of the given determination matter by one of the plurality of determination methods for corresponding one of the plurality of reliability prediction models is associated with, as correct data, (ii) information indicating whether or not a result of the determination is correct.
2 . The information processing device as set forth in claim 1 , wherein the target data is an image of an inspection target;
wherein the given determination matter is presence or absence of an abnormal portion in the inspection target;
wherein each of the plurality of determination methods includes a determination method for determining presence or absence of the abnormal portion with use of a generated image generated by inputting the image that is the target data into a generative model; and
wherein the generative model is constructed by machine learning that uses, as training data, an image of an inspection target not having the abnormal portion, the generative model being constructed to generate a new image having a similar feature to an image input into the generative model.
3 . The information processing device as set forth in claim 2 , wherein the plurality of determination methods includes a determination method for analyzing pixel values in the target data, which is the image of the inspection target, so as to identify an inspection target portion in the target data and for determining presence or absence of the abnormal portion in accordance with pixel values in the inspection target portion thus identified.
4 . The information processing device as set forth in claim 3 , wherein the target data is an ultrasonic testing image which is an image of an echo of an ultrasonic wave propagated in the inspection target;
wherein in the determination method for determining presence or absence of the abnormal portion in accordance with the pixel values in the inspection target portion, the processor identifies, as the inspection target portion, an area sandwiched between two peripheral echo areas in each of which an echo coming from a periphery of the inspection target portion appears repeatedly, and determines presence or absence of the abnormal portion in accordance with whether or not the inspection target portion thus identified includes an area constituted by pixel values each being not less than a threshold; and
wherein the processor further carries out a thickness calculating step of calculating a thickness of the inspection target portion in accordance with a distance between the two peripheral echo areas.
5 . A determination method being executed by one or more information processing devices, comprising:
a reliability determining step of determining, using single target data, reliabilities which are indices indicating degrees of certainties of determination results obtained, from the single target data, by determining a given determination matter by a plurality of determination methods that are different from each other; and
a comprehensive determining step of determining the given determination matter with use of the determination results and the reliabilities determined in the reliability determining step,
wherein, in the reliability determining step, the one or more information processing devices determine the plurality of reliabilities, in accordance with output values obtained by inputting the target data into a plurality of reliability prediction models respectively corresponding to the plurality of determination methods; and
wherein each of the plurality of reliability prediction models is constructed by machine learning that uses training data in which (i) data used for determination of the given determination matter by one of the plurality of determination methods for corresponding one of the plurality of reliability prediction models is associated with, as correct data, (ii) information indicating whether or not a result of the determination is correct.
6 . A non-transitory computer readable medium storing an information processing program configured to cause a computer to function as an information processing device recited in claim 1 , the information processing program causing the computer to carry out the reliability determining step and the comprehensive determination step.