Data processing method and learning method
A data processing method includes an input step S 1 of inputting measurement data into a neural network, an estimation step S 2 of generating estimation data from the measurement data, a restoration step S 3 of generating restoration data from the estimation data, and a calculation step S 4 of calculating a confidence level of the estimation data, based on the measurement data and the restoration data. The neural network is a trained model, the measurement data is data obtained by measuring light transmitted through an object, the estimation data is data of a three-dimensional optical characteristic of the object estimated from the measurement data, and the three-dimensional optical characteristic is a refractive index distribution or an absorptance distribution. In the estimation, the neural network is used, in the restoration, forward propagation operations are performed on the estimation data, and in the forward propagation operations, wavefronts passing through the interior of the object estimated from the measurement data are sequentially obtained in a direction in which light travels.
1 . A data processing method comprising:
an input step of inputting measurement data into a neural network;
an estimation step of generating estimation data from the measurement data;
a restoration step of generating restoration data from the estimation data; and
a calculation step of calculating a confidence level of the estimation data, based on the measurement data and the restoration data, wherein
the neural network is a trained model,
the measurement data is data obtained by measuring light transmitted through an object,
the estimation data is data of a three-dimensional optical characteristic of the object estimated from the measurement data,
the three-dimensional optical characteristic is a refractive index distribution or an absorptance distribution,
in the estimation, the neural network is used,
in the restoration, forward propagation operations are performed on the estimation data, and
in the forward propagation operations, wavefronts passing through interior of the object estimated from the measurement data are sequentially obtained in a direction in which light travels.
2 . The data processing method according to claim 1 , wherein
a difference between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on the difference.
3 . The data processing method according to claim 1 , wherein
a correlation between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on the correlation.
4 . The data processing method according to claim 1 , wherein
a sum of squares of a difference between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on a magnitude of the sum of squares of the difference.
5 . The data processing method according to claim 1 , comprising:
a first neural network; and a second neural network, wherein
the first neural network is the neural network,
the second neural network is a trained model, and
the confidence level is calculated using the second neural network.
6 . The data processing method according to claim 5 , wherein
the second neural network learns using a first training data set group and a second training data set group,
the first training data set group includes a plurality of first training data sets,
the first training data sets each include first data, first corrected data, and teaching data indicating true between true and false,
the second training data set group includes a plurality of second training data sets,
the second training data sets each include the first data, second corrected data, and teaching data indicating false between true and false,
the first corrected data is data obtained by performing a correction process on the first data,
the second corrected data is data obtained by performing a correction process on second data,
the second data is different from the first data, and
the first data and the second data are data obtained by measuring light transmitted through the object, or data generated by performing forward propagation operations on an object model that models a three-dimensional optical characteristic of the object.
7 . The data processing method according to claim 1 , further comprising a presentation step of presenting the confidence level.
8 . A data processing method comprising:
an input step of inputting measurement data into a neural network;
an estimation step of generating estimation data from the measurement data;
a restoration step of generating restoration data from the estimation data;
a calculation step of calculating a confidence level of the estimation data, based on the measurement data and the restoration data; and
a learning step of learning by the neural network with a quantity inversely proportional to the confidence level as a loss, wherein
the measurement data is data obtained by measuring light transmitted through an object,
the estimation data is data of a three-dimensional optical characteristic of the object estimated from the measurement data,
the three-dimensional optical characteristic is a refractive index distribution or an absorptance distribution,
in the estimation, the neural network is used,
in the restoration, forward propagation operations are performed on the estimation data, and
in the forward propagation operations, wavefronts passing through interior of the object estimated from the measurement data are sequentially obtained in a direction in which light travels.
9 . The data processing method according to claim 8 , wherein
a difference between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on the difference.
10 . The data processing method according to claim 8 , wherein
a correlation between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on the correlation.
11 . The data processing method according to claim 8 , wherein
a sum of squares of a difference between the measurement data and the restoration data is calculated, and
the confidence level is calculated based on a magnitude of the sum of squares of the difference.
12 . The data processing method according to claim 8 , comprising:
a first neural network; and a second neural network, wherein
the first neural network is the neural network,
the second neural network is a trained model, and
the confidence level is calculated using the second neural network.
13 . The data processing method according to claim 12 , wherein
the second neural network learns using a first training data set group and a second training data set group,
the first training data set group includes a plurality of first training data sets,
the first training data sets each include first data, first corrected data, and teaching data indicating true between true and false,
the second training data set group includes a plurality of second training data sets,
the second training data sets each include the first data, second corrected data, and teaching data indicating false between true and false,
the first corrected data is data obtained by performing a correction process on the first data,
the second corrected data is data obtained by performing a correction process on second data,
the second data is different from the first data, and
the first data and the second data are data obtained by measuring light transmitted through the object, or data generated by performing forward propagation operations on an object model that models a three-dimensional optical characteristic of the object.
14 . The data processing method according to claim 8 , further comprising a presentation step of presenting the confidence level.
15 . A learning method for a neural network to calculate a confidence level of estimation data, wherein
the confidence level of the estimation data is calculated based on measurement data and restoration data,
the measurement data is data obtained by measuring light transmitted through an object,
the estimation data is data of a three-dimensional optical characteristic of the object estimated from the measurement data,
the three-dimensional optical characteristic is a refractive index distribution or an absorptance distribution,
the restoration data is data generated by performing forward propagation operations on the estimation data,
in the forward propagation operations, wavefronts passing through interior of the object estimated from the measurement data are sequentially obtained in a direction in which light travels,
the learning method includes:
a first learning step of learning using a first training data set; and
a second learning step of learning using a second training data set,
the first learning step and the second learning step are repeatedly performed,
the first training data set includes first data, first corrected data, and teaching data indicating true between true and false,
the second training data set includes the first data, second corrected data, and the teaching data indicating false between true and false,
the first corrected data is data obtained by performing a correction process on the first data,
the second corrected data is data obtained by performing a correction process on second data,
the second data is different from the first data, and
the first data and the second data are data obtained by measuring light transmitted through the object, or data generated by performing forward propagation operations on an object model that models a three-dimensional optical characteristic of the object.
16 . The learning method for a neural network according to claim 15 , wherein the correction process performed on the first data and the correction process performed on the second data include at least one process among a deforming process, a rotating process, and a noise adding process.