JP2024088356A - Eyeglass lens design system - Google Patents

Abstract

An eyeglass lens design system suitable for receiving an order for the design of an eyeglass lens via a computer network is provided.
[Solution] One aspect of the present invention provides a spectacle lens design system for designing a spectacle lens. The spectacle lens design system includes a parameter prediction unit that predicts parameters for optically calculating the shape of the rear surface of the spectacle lens from a data group including spectacle lens data related to the spectacle lens using a parameter prediction model. The spectacle lens design system includes a rear surface shape calculation unit that calculates the shape of the rear surface of the spectacle lens. The parameter prediction model is a prediction model trained by machine learning using the data group and parameters as training data. The rear surface shape calculation unit calculates the shape of the rear surface of the spectacle lens using the parameters predicted by the parameter prediction unit.
[Selected figure] Figure 4

Description

The present invention relates to an eyeglass lens design system.

A spectacle lens manufacturer may receive an order for the design of a spectacle lens via a computer network. For example, a staff member at an eyeglass store uses a terminal at the eyeglass store to send data on the eyeglass lens to a server at the eyeglass lens manufacturer. Upon receiving the data, the eyeglass lens manufacturer's server designs the eyeglass lens (see, for example, Patent Document 1) and sends data showing the design results to the terminal at the eyeglass store.

The invention described in Patent Document 1 relates to a method for designing eyeglass lenses. The invention described in Patent Document 1 makes it possible to provide eyeglass lenses that meet the individual needs of each wearer with regard to vision near the eyepoint.

JP 2014-085575 A

One aspect of the present invention provides a spectacle lens design system for designing a spectacle lens. The spectacle lens design system includes a parameter prediction unit that predicts parameters for optically calculating the shape of the rear surface of the spectacle lens from a data group including spectacle lens data related to the spectacle lens using a parameter prediction model. The spectacle lens design system includes a rear surface shape calculation unit that calculates the shape of the rear surface of the spectacle lens. The parameter prediction model is a prediction model that has been machine-learned using the data group and the parameters as training data. The rear surface shape calculation unit calculates the shape of the rear surface of the spectacle lens using the parameters predicted by the parameter prediction unit.

Note that the above summary of the invention does not list all of the features of the present invention. Also, subcombinations of the features of the present invention may be inventions.

FIG. 2 is a diagram illustrating an example of a usage environment of a design server 100. FIG. 2 illustrates an example of a configuration of a design server 100. FIG. 13 is a diagram illustrating an example of a data group. FIG. 11 is a diagram illustrating an example of processing by the design server 100. FIG. 13 is a diagram showing the results of verifying the prediction accuracy by a parameter prediction model. FIG. 13 is a diagram showing the results of verifying the prediction time according to the parameter prediction model.

The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

Figure 1 is a diagram showing an example of a usage environment of the design server 100. The design server 100 is a server that designs spectacle lenses. Spectacle lenses are ophthalmic lenses that are worn in front of the eye without contacting the eyeball. Ophthalmic lenses are lenses used to measure, correct, or protect the eye, or to change the appearance. The design server 100 is an example of a design system.

The design server 100 is communicatively connected to the eyeglass store terminal OT via a computer network CN. The computer network CN is a communication network that connects multiple computers via signal lines or wirelessly, allowing them to send and receive data to and from each other. The eyeglass store terminal OT is provided in the eyeglass store OS, and is a terminal used by the staff of the eyeglass store OS.

Figure 2 is a diagram showing an example of the configuration of the design server 100. The design server 100 includes a CPU 110 (central processing unit), a main memory 120, an input/output interface 130, a communication device 140, and storage 150.

The CPU 110 is a device that controls other devices and circuits and performs data calculations, etc. The main memory 120 is one of the storage devices that stores data and programs and is directly connected to the CPU 110 via electrical wiring on a board, etc. The input/output interface 130 is a hardware interface that connects multiple devices for communication. The communication device 140 is a device that transmits and receives electricity, radio waves, and light to communicate with other devices. The storage 150 is a device that permanently stores data.

The CPU 110 has programs installed and functions as a data group receiving unit SM1, a parameter prediction unit SM2, a processing time prediction unit SM3, a change data prediction unit SM4, a rear surface shape calculation unit SM5, and a shape data transmission unit SM6.

The data group receiving unit SM1 is a software module that receives a data group from the eyeglass store terminal OT. The data group is a collection of data that includes at least data related to eyeglass lenses. The data group may include data related to the prescription of a wearer of the eyeglass lenses. The data group may include data related to eyeglass frames. The data included in the data group received by the data group receiving unit SM1 is stored in storage 150.

Figure 3 is a diagram showing an example of a data group. The data group shown in Figure 3 includes eyeglass lens data D1, prescription data D2, and eyeglass frame data D3.

The eyeglass lens data D1 is data related to the eyeglass lens desired by the wearer. The eyeglass lens data D1 includes base curve data D11, lens outer diameter data D12, center thickness data D13, front vertex power data D14, etc.

The base curve data D11 is data that indicates the base curve. The base curve is usually the surface power of the front surface. The front surface is the surface of the spectacle lens that is farther from the wearer's eye. The surface power is the local ability of the finished surface to change the vergence of a bundle of rays that are incident on the surface.

Lens outer diameter data D12 is data that indicates the outer diameter of the eyeglass lens.

The center thickness data D13 is data indicating the center thickness. The center thickness is the thickness at the reference point of a blank or spectacle lens measured perpendicular to the front surface. A blank is an optical material part with one completed surface for making a spectacle lens. An optical material is a transparent material suitable for manufacturing optical parts. The measurement reference point is a point on the front surface of a finished lens or the finished surface of a blank to which the verification power of the corresponding area is applied. A finished lens is a spectacle lens with a final optical surface on both sides. The verification power is the refractive power of a spectacle lens specially calculated and provided by a spectacle lens manufacturer for verification on an instrument.

The front vertex power data D14 is data indicating the front vertex power. The front vertex power is the reciprocal of the distance from the vertex of the front surface of the spectacle lens to the focal point in the paraxial region.

The prescription data D2 is data related to the prescription of the eyeglass lens wearer. The prescription data D2 includes spherical power data D21, cylindrical power data D22, cylindrical axis angle data D23, and addition power data D24.

The spherical power data D21 is data indicating the spherical power. In the case of a spherical-power lens, the spherical power is the value of the back vertex power of the spherical-power lens. In the case of an astigmatic power lens, the spherical power is the value of the back vertex power of the principal meridian selected as a reference from the two principal meridians of the astigmatic power lens. A spherical power lens is a spectacle lens in which a paraxial bundle of parallel light brings a single focal point. The back vertex power is the reciprocal of the distance from the rear vertex of the spectacle lens to the focal point in the paraxial region.

The cylindrical power data D22 is data indicating the cylindrical power. The cylindrical power is the algebraic difference between the two principal meridian powers obtained by subtracting the principal meridian power selected as a reference from the other principal meridian power. The principal meridian power is the rear vertex power of one of the two principal meridians of the cylindrical power lens.

The astigmatic axis angle data D23 is data indicating the angle of the astigmatic axis. The astigmatic axis is the direction of the principal meridian of a spectacle lens having a vertex power selected as a reference. The vertex power is the reciprocal of the distance from the vertex to the focal point in the paraxial region. The vertex is the intersection of the optical axis of the spectacle lens with the surface of the spectacle lens. The optical axis is a straight line connecting the centers of curvature of both surfaces of the spectacle lens. The principal meridian is one of the two mutually perpendicular meridians of an astigmatic-power lens that is parallel to the two focal lines. An astigmatism lens is a spectacle lens in which a paraxial beam of parallel light is split into two, resulting in focal lines that are perpendicular to each other, and which has vertex power only in the two principal meridians. The meridians are each plane that contains the optical axis of the spectacle lens.

The addition power data D24 is data indicating the addition power. The addition power is the difference between the vertex power of the near portion and the vertex power of the distance portion (3.15.1). The near portion is a portion having a refractive power for near vision distance, such as a multifocal lens. A multifocal lens is a spectacle lens designed to provide two or more different focal powers that are visually separated. The near vision distance is the working distance when the wearer's habitual near working plane is the object plane. The working distance is the distance from the reference plane to the object plane. The distance portion is a portion having a refractive power for distance vision, such as a multifocal lens.

The eyeglass frame data D3 is data related to the eyeglass frame desired by the wearer. The eyeglass frame data D3 includes the bending angle data D31 when wearing, the forward tilt angle data D32 when wearing, the vertex distance data D33, the frame shape data D34, the visual point data D35, and the interpupillary distance data D36.

The as-worn face form data D31 is data indicating the as-worn face form angle. The as-worn face form angle is the horizontal angle between the perpendicular line to the reference line passing through the apex of the nose-side and ear-side rim grooves of the spectacle frame and the primary direction in a horizontal plane including the primary direction. The primary direction is the direction of the line of sight toward an object at infinity, measured with the head and torso in a habitual position when looking straight ahead without visual aids, which is usually considered to be the horizontal direction. The line of sight is a first light path from a point of interest in the object-side space toward the center of the entrance pupil of the eye, and a second light path that is continuous with the first light path and extends from the center of the exit pupil toward a fixed point on the retina.

The as-worn pantoscopic angle data D32 is data indicating the as-worn pantoscopic angle. The as-worn pantoscopic angle is the vertical angle between the horizon and a perpendicular line to a reference line passing through the apex of the grooves on the upper and lower rims of the eyeglass frame in a vertical plane including the first eye position direction.

The vertex distance data D33 is data indicating the vertex distance. The vertex distance is the horizontal distance between the back surface of the spectacle lens and the corneal apex measured when the eye is in the primary position. The primary position is the eye position when looking in the primary position direction. The back surface is the surface of the spectacle lens that is closer to the wearing eye.

Frame shape data D34 is data that indicates the shape of the eyeglass frame.

Visual point data D35 is data that indicates the visual point. The visual point is the intersection of the line of sight and the rear surface of the eyeglass lens. The visual point is also called the eye point.

The interpupillary distance data D36 is data indicating the interpupillary distance. The interpupillary distance is the distance between the centers of the two eye openings when both eyes are in the first eye position.

The parameter prediction unit SM2 is a software module that predicts parameters for calculating the shape of the rear surface of the eyeglass lens from a group of data using a parameter prediction model. The parameter prediction model is a prediction model generated by supervised machine learning using a known machine learning algorithm such as a neural network. The parameter prediction model is generated, for example, by learning the relationship between a group of previously acquired data and the correct answer of the parameters for calculating the shape of the rear surface of the eyeglass lens as supervised data.

Storage 150 stores the first parameter prediction model PM1 to the fifth parameter prediction model PM5 as parameter prediction models.

The first parameter prediction model PM1 is generated by supervised machine learning, for example, in which a data group including eyeglass lens data D1 in which the base curve of the eyeglass lens is 8.00 is used as training data. When the base curve of the eyeglass lens selected based on the prescription of the wearer is 8.00, the parameter prediction unit SM2 predicts parameters from the data group using the first parameter prediction model PM1. The base curve of the eyeglass lens is set so that multiple types of values can be selected for each prescription range predetermined by the eyeglass lens manufacturer. As the base curve of the eyeglass lens suitable for the wearer, one of multiple values is selected based on the prescription of the wearer. The base curve of 8.00 is an example of a base curve of the first value.

Here, when the base curve is 8.00, the spherical refractive power can take values ranging from +8.00 to +7.25 at intervals of 0.25, for example. The cylindrical refractive power can take values ranging from -0.00 to -6.00 at intervals of 0.25, for example. If the first parameter prediction model PM1 is generated using a data group of all combinations of possible values of the spherical refractive power and possible values of the cylindrical refractive power as training data, there is a possibility of overlearning. Therefore, the first parameter prediction model PM1 is generated using a data group of combinations of some of the possible values of the spherical refractive power and some of the possible values of the cylindrical refractive power as training data. For example, the first parameter prediction model PM1 is generated using a data group of combinations of values of the spherical refractive power with a decimal point of 0 and values of the cylindrical refractive power with a decimal point of 0 as training data.

The addition power can take values from 0.50 to 4.00 in increments of 0.25, for example. If the first parameter prediction model PM1 is generated using a data set of all combinations of possible values of the spherical power and possible values of the addition power as training data, there is a possibility of overlearning. Therefore, the first parameter prediction model PM1 is generated using a data set of combinations of some of the possible values of the spherical power and some of the possible values of the addition power as training data. For example, the first parameter prediction model PM1 is generated using a data set of combinations of values of the spherical power with a decimal point of 0 and values of the addition power with a decimal point of 0 as training data.

The second parameter prediction model PM2 is generated by supervised machine learning, for example, in which a data group including eyeglass lens data D1 in which the base curve of the eyeglass lens is 6.75 is used as training data. When the base curve of the eyeglass lens selected based on the prescription of the wearer is 6.75, the parameter prediction unit SM2 predicts parameters from the data group using the second parameter prediction model PM2. The base curve of 6.75 is an example of a base curve of the second value.

Here, when the base curve is 6.75, the spherical refractive power can take values ranging from +7.00 to +5.25 at intervals of 0.25, for example. The cylindrical refractive power can take values ranging from -0.00 to -6.00 at intervals of 0.25, for example. If the second parameter prediction model PM2 is generated using a data group of all combinations of possible values of the spherical refractive power and possible values of the cylindrical refractive power as training data, there is a possibility of overlearning. Therefore, the second parameter prediction model PM2 is generated using a data group of combinations of some of the possible values of the spherical refractive power and some of the possible values of the cylindrical refractive power as training data. For example, the second parameter prediction model PM2 is generated using a data group of combinations of values of the spherical refractive power with a decimal point of 0 and values of the cylindrical refractive power with a decimal point of 0 as training data.

The addition power can take values from 0.50 to 4.00 in increments of 0.25, for example. If the second parameter prediction model PM2 is generated using as training data a data group of all combinations of possible values of the spherical power and possible values of the addition power, there is a possibility of overlearning. Therefore, the second parameter prediction model PM2 is generated using as training data a data group of combinations of some of the possible values of the spherical power and some of the possible values of the addition power. For example, the second parameter prediction model PM2 is generated using as training data a data group of combinations of values of the spherical power with a decimal point of 0 and values of the addition power with a decimal point of 0.

The third parameter prediction model PM3 is generated by supervised machine learning, in which a data group including eyeglass lens data D1 in which the base curve of the eyeglass lens is 6.00 is used as training data. When the base curve of the eyeglass lens selected based on the wearer's prescription is 6.00, the parameter prediction unit SM2 predicts parameters from the data group using the third parameter prediction model PM3.

Here, when the base curve is 6.00, the spherical refractive power can take values ranging from +5.00 to +4.00 at intervals of 0.25, for example. The cylindrical refractive power can take values ranging from -0.00 to -6.00 at intervals of 0.25, for example. If the third parameter prediction model PM3 is generated using a data group of all combinations of possible values of the spherical refractive power and possible values of the cylindrical refractive power as training data, there is a possibility of overlearning. Therefore, the third parameter prediction model PM3 is generated using a data group of combinations of some of the possible values of the spherical refractive power and some of the possible values of the cylindrical refractive power as training data. For example, the third parameter prediction model PM3 is generated using a data group of combinations of values of the spherical refractive power with a decimal point of 0 and values of the cylindrical refractive power with a decimal point of 0 as training data.

The addition power can take values from 0.50 to 4.00 in increments of 0.25, for example. If the third parameter prediction model PM3 is generated using a data set of all combinations of possible values of the spherical power and possible values of the addition power as training data, there is a possibility of overlearning. Therefore, the third parameter prediction model PM3 is generated using a data set of combinations of some of the possible values of the spherical power and some of the possible values of the addition power as training data. For example, the third parameter prediction model PM3 is generated using a data set of combinations of values of the spherical power with a decimal point of 0 and values of the addition power with a decimal point of 0 as training data.

The fourth parameter prediction model PM4 is generated by supervised machine learning, in which a data group including eyeglass lens data D1 in which the base curve of the eyeglass lens is 2.75 is used as training data. When the base curve of the eyeglass lens selected based on the prescription of the wearer is 2.75, the parameter prediction unit SM2 predicts parameters from the data group using the fourth parameter prediction model PM4.

Here, when the base curve is 2.75, the spherical refractive power can take values from +3.75 to +2.00 in increments of 0.25, for example. When the spherical refractive power is +3.75, the cylindrical refractive power can take values from -0.00 to -5.75 in increments of 0.25. When the spherical refractive power is reduced by 0.25, for example, the range of cylindrical refractive power from -0.00 narrows by 0.25. When the spherical refractive power is +2.00, for example, the cylindrical refractive power can take values from -0.00 to -4.00 in increments of 0.25. When the fourth parameter prediction model PM4 is generated using a data group of all combinations of values that the spherical refractive power and values that the cylindrical refractive power can take as training data, there is a possibility of overlearning. Therefore, the fourth parameter prediction model PM4 is generated using a data group of combinations of some of the possible values of the spherical refractive power and some of the possible values of the cylindrical refractive power as training data. For example, the fourth parameter prediction model PM4 is generated using a data group of combinations of values of the spherical refractive power with a decimal point of 0 and values of the cylindrical refractive power with a decimal point of 0 as training data.

The addition power can take values from 0.50 to 4.00 in increments of 0.25, for example. If the fourth parameter prediction model PM4 is generated using a data set of all combinations of possible values of the spherical power and possible values of the addition power as training data, there is a possibility of overlearning. Therefore, the fourth parameter prediction model PM4 is generated using a data set of combinations of some of the possible values of the spherical power and some of the possible values of the addition power as training data. For example, the fourth parameter prediction model PM4 is generated using a data set of combinations of values of the spherical power with a decimal point of 0 and values of the addition power with a decimal point of 0 as training data.

The fifth parameter prediction model PM5 is generated by supervised machine learning, in which a data group including eyeglass lens data D1 in which the base curve of the eyeglass lens is 1.75 is used as training data. When the base curve of the eyeglass lens selected based on the wearer's prescription is 1.75, the parameter prediction unit SM2 predicts parameters from the data group using the fifth parameter prediction model PM5.

Here, when the base curve is 1.75, the spherical refractive power can take a value of +3.75 to -8.00 in 0.25 increments, for example. The cylindrical refractive power can take a value of -6.00 when the spherical refractive power is +3.75, for example. The cylindrical refractive power narrows in range up to +6.00 by 0.25 increments each time the spherical refractive power decreases by 0.25 to +2.00, for example. The cylindrical refractive power can take a value of -4.25 to -6.00 in 0.25 increments when the spherical refractive power is +2.00, for example. The cylindrical refractive power can take a value of -0.00 to -6.00 in 0.25 increments when the spherical refractive power is +1.75 to -2.00, for example. For example, the range of the astigmatic power from -0.00 narrows by 0.25 each time the spherical power decreases from -2.25 by 0.25. For example, when the spherical power is -8.00, the astigmatic power can take a value of -0.00. If the fifth parameter prediction model PM5 is generated using a data group of all combinations of values that the spherical power can take and values that the astigmatic power can take as training data, there is a possibility of overlearning. Therefore, the fifth parameter prediction model PM5 is generated using a data group of combinations of some values of the values that the spherical power can take and some values of the values that the astigmatic power can take as training data. For example, the fifth parameter prediction model PM5 is generated using a data group of combinations of values of the spherical power that are 0 after the decimal point and values of the astigmatic power that are 0 after the decimal point as training data.

The addition power can take values from 0.50 to 4.00 in increments of 0.25, for example. If the fifth parameter prediction model PM5 is generated using a data set of all combinations of possible values of the spherical power and possible values of the addition power as training data, there is a possibility of overlearning. Therefore, the fifth parameter prediction model PM5 is generated using a data set of combinations of some of the possible values of the spherical power and some of the possible values of the addition power as training data. For example, the fifth parameter prediction model PM5 is generated using a data set of combinations of values of the spherical power with a decimal point of 0 and values of the addition power with a decimal point of 0 as training data.

The processing time prediction unit SM3 is a software module that uses the processing time prediction model PM6 to predict whether the shape of the rear surface of the eyeglass lens can be determined within a specified time from the above-mentioned parameters. The processing time prediction model PM6 is a prediction model generated by supervised machine learning using a known machine learning algorithm such as a neural network. The processing time prediction model PM6 is generated by learning the relationship between the parameters for calculating the shape of the rear surface of the eyeglass lens and actual data on whether the shape of the rear surface of the eyeglass lens could be determined within a specified time as supervised data. The processing time prediction model PM6 is stored in the storage 150.

If the parameter prediction unit SM2 predicts that the shape of the rear surface of the eyeglass lens cannot be determined within a specified time, it changes some of the data in the data group and re-predicts the parameters.

The shape data transmission unit SM6 is a software module that transmits shape data D5 relating to the shape of the rear surface of the eyeglass lens to the eyeglass store terminal OT. The shape data D5 is data indicating the shape of the rear surface of the eyeglass lens calculated by the rear surface shape calculation unit SM5, and is stored in the storage 150.

Here, when the design server 100 receives the data group from the optician's terminal OT, it calculates the shape of the rear surface of the eyeglass lens and transmits shape data D5 to the optician's terminal OT. If the optician's terminal OT is unable to receive the shape data D5 even after a preset timeout period has elapsed since transmitting the data group, it terminates communication.

The processing time prediction unit SM3 predicts whether the shape of the rear surface of the eyeglass lens can be determined within the remaining time of the timeout period from when the data group is received until when the shape data D5 is transmitted.

The change data prediction unit SM4 is a software module that uses the change data prediction model PM7 to predict which data in the data group should be changed based on the remaining time in the timeout period. The change data prediction model PM7 is a prediction model generated by supervised machine learning using a known machine learning algorithm such as a neural network. The change data prediction model PM7 is generated by learning the relationship between performance data on whether the shape of the rear surface of the eyeglass lens was determined within a specified time and a data group including at least eyeglass lens data D1 as teacher data. The data group may include prescription data D2. The data group may include eyeglass frame data D3. The change data prediction model PM7 is stored in storage 150.

The parameter prediction unit SM2 modifies the data predicted by the change data prediction unit SM4 and re-predicts the parameters.

The rear surface shape calculation unit SM5 is a software module that calculates the shape of the rear surface of the eyeglass lens. The rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens using parameters predicted by the parameter prediction unit SM2. The rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens by ray tracing calculation and wavefront calculation. Ray tracing calculation is a calculation method that is the basis of calculations in eyeglass lens design and optical system evaluation. More specifically, ray tracing calculation is a method of repeatedly refracting light rays according to Snell's law at the boundary between the eyeglass lens and air through which the light rays pass, while taking into account the surface shape of the eyeglass lens, and obtaining the image plane position where the light rays ultimately reach. The position at which the light rays from an object point enter the eyeglass lens can be calculated from the position, direction, and information on the eyeglass lens surface. The wavefront calculation is a method of calculating the optical path length from the object point to a reference plane centered on the image.

Figure 4 shows an example of processing by the design server 100.

When the staff of the eyeglass store OS prescribes a pair of glasses for the wearer, they input prescription data D2 relating to the prescription for the wearer into the eyeglass store terminal OT. When the staff of the eyeglass store OS decides on the eyeglass frame desired by the wearer, they input eyeglass frame data D3 relating to the eyeglass frame desired by the wearer into the eyeglass store terminal OT. When the staff of the eyeglass store OS decides on the eyeglass lenses desired by the wearer, they input eyeglass lens data D1 relating to the eyeglass lenses desired by the wearer into the eyeglass store terminal OT. Then, in order to have the shape of the eyeglass frame desired by the wearer designed, the staff of the eyeglass store OS sends a data group including the data input into the eyeglass store terminal OT to the design server 100 using the eyeglass store terminal OT. After sending the data group, the eyeglass store terminal OT goes into a standby state, waiting for a response from the design server 100 until a preset timeout time has elapsed.

The data group receiving unit SM1 receives the data group transmitted from the eyeglass store terminal OT (S101). In S101, when the data group receiving unit SM1 receives the data group, it stores the eyeglass lens data D1, prescription data D2, and eyeglass frame data D3 contained in the data group in the storage 150.

When the data group is received, the parameter prediction unit SM2 predicts parameters for optically calculating the shape of the rear surface of the eyeglass lens from the data group using the parameter prediction model (S102). In S102, the parameter prediction unit SM2 predicts parameters for optically calculating the shape of the rear surface of the eyeglass lens using the parameter prediction model. The parameter prediction unit SM2 varies the parameter prediction model used to predict the parameters depending on the base curve of the eyeglass lens selected based on the wearer's prescription. As described above, when the base curve selected based on the wearer's prescription is 8.00, the parameter prediction unit SM2 predicts parameters from the data group using the first parameter prediction model PM1. When the base curve selected based on the wearer's prescription is 6.75, the parameter prediction unit SM2 predicts parameters from the data group using the second parameter prediction model PM2. When the base curve selected based on the wearer's prescription is 6.00, the parameter prediction unit SM2 predicts parameters from the data group using the third parameter prediction model PM3. When the base curve selected based on the wearer's prescription is 2.75, the parameter prediction unit SM2 predicts the parameters from the data group using the fourth parameter prediction model PM4. When the base curve selected based on the wearer's prescription is 1.75, the parameter prediction unit SM2 predicts the parameters from the data group using the fifth parameter prediction model PM5. When the parameter prediction unit SM2 predicts the parameters, it stores parameter data D4 indicating the parameters in storage 150.

When the parameters are predicted, the processing time prediction unit SM3 predicts whether the shape of the rear surface of the eyeglass lens can be determined within the remaining time of the timeout time from the parameters (S103). In S103, the processing time prediction unit SM3 predicts whether the shape of the rear surface of the eyeglass lens can be determined within the remaining time of the timeout time using the processing time prediction model PM6. As described above, the rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens using the parameters predicted by the parameter prediction unit SM2. The shape of the rear surface of the eyeglass lens calculated by the rear surface shape calculation unit SM5 is an optically ideal shape for the eyeglass lens of the wearer. However, the shape of the rear surface of the eyeglass lens calculated by the rear surface shape calculation unit SM5 may be a shape that cannot be processed by a device that processes the rear surface of the eyeglass lens. As described above, the processing time prediction model PM6 is generated as teacher data on the relationship between the parameters for calculating the shape of the rear surface of the eyeglass lens and the performance data on whether the shape of the rear surface of the eyeglass lens was determined within a specified time. The performance data may be data indicating that the shape of the rear surface of the eyeglass lens was determined within a predetermined time when the shape of the rear surface of the eyeglass lens calculated from the parameters is a shape that can be processed. Also, the performance data may be data indicating that the shape of the rear surface of the eyeglass lens could not be determined within a predetermined time when the shape of the rear surface of the eyeglass lens calculated from the parameters is a shape that cannot be processed.

If it is predicted that the decision cannot be made within the remaining time (S104; NO), the change data prediction unit SM4 predicts which data in the data group should be changed from the remaining time (S105). In S105, the change data prediction unit SM4 predicts which data in the data group should be changed using the change data prediction model PM7. Here, the staff of the eyeglass store OS transmits the data group to the design server 100, for example, to confirm the optimal center thickness for the eyeglass lens desired by the wearer. For example, if the center thickness data D13 is included in the data group, once the shape of the rear surface of the eyeglass lens that can be processed is determined, the staff of the eyeglass store OS can determine that the center thickness indicated by the center thickness data D13 is appropriate. In view of the demand to confirm the center thickness of the eyeglass lens, it is sufficient that the change data prediction model PM7 is learned so as to easily predict that the center thickness data D13 should be changed.

When data that needs to be changed is predicted, the parameter prediction unit SM2 changes the data predicted by the change data prediction unit SM4 and re-predicts the parameters (S106). In S106, the parameter prediction unit SM2 uses the parameter prediction model to re-predict the parameters for optically calculating the shape of the rear surface of the eyeglass lens.

If it is predicted that the shape can be determined within the remaining time (S104: YES), or after the processing of S106 has been executed, the rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens (S107). In S107, the rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens using the parameters predicted by the parameter prediction unit SM2. The rear surface shape calculation unit SM5 calculates the shape of the rear surface of the eyeglass lens by ray tracing calculation and wavefront calculation. After calculating the shape of the rear surface of the eyeglass lens, the rear surface shape calculation unit SM5 stores shape data D5 indicating the shape in storage 150.

Once the shape of the rear surface of the eyeglass lens has been calculated, the shape data transmission unit SM6 transmits the shape data D5 to the eyeglass store terminal OT that transmitted the data group received in S101 (S108).

Figure 5 shows the results of verifying the prediction accuracy of a parameter prediction model. The verification results shown in Figure 5 are the results of verifying the prediction accuracy using three types of parameter prediction models with different learned spherical refractive powers. The prediction accuracy is the error between the predicted parameter value and the correct answer when the base curve of the eyeglass lens selected based on the wearer's prescription is 8.00. The correct answer was calculated using a known method with high accuracy.

The first parameter prediction model used in the verification was generated by learning the relationship between a data group including spherical refractive power data D21 with a spherical refractive power of +8.00 and the correct parameter answer as training data. Note that a spherical refractive power of +8.00 corresponds to a base curve of 8.00. The number of patterns learned was 1671. When the base curve of the eyeglass lens selected based on the wearer's prescription is 8.00, the prediction accuracy of the parameters using the first parameter prediction model was -2% to +1%.

The second parameter prediction model used in the verification was generated by learning the relationship between a data group including spherical refractive power data D21 with spherical refractive powers of +8.00 to +6.50 and the correct parameter answers as training data. Note that spherical refractive powers of +8.00 to +6.50 correspond to base curves of 8.00 to 6.75. The number of patterns learned was 3341. When the base curve of the eyeglass lens selected based on the wearer's prescription is 8.00, the parameter prediction accuracy using the second parameter prediction model was -2% to +2%.

The third parameter prediction model used in the verification was generated by learning the relationship between a data group including spherical refractive power data D21 with spherical refractive power of +8.00 to -8.00 and the correct parameter answer as training data. Note that spherical refractive power of +8.00 to -8.00 corresponds to a base curve of 8.00 to 1.75. The number of patterns learned was 17035. When the base curve of the eyeglass lens selected based on the wearer's prescription is 8.00, the prediction accuracy of the parameters using the third parameter prediction model was -2% to +3%.

According to the verification results shown in Figure 5, the prediction accuracy of the second parameter prediction model is higher than the prediction accuracy of the third parameter prediction model. In other words, it can be seen that the prediction accuracy of the parameter prediction model can be improved by using a prediction model that learns from data with a narrow learning range that includes the spherical refractive power corresponding to the target base curve.

According to the verification results shown in Figure 5, the prediction accuracy of the first parameter prediction model is higher than the prediction accuracy of the second parameter prediction model. In other words, it can be seen that the prediction accuracy of the parameter prediction model can be improved by using a prediction model that learns only data on the spherical refractive power corresponding to the target base curve.

Figure 6 shows the results of verifying the prediction time using a parameter prediction model. The verification results shown in Figure 6 are the results of verifying the prediction time using the same three types of parameter prediction models as the prediction models used in the verification shown in Figure 5. The prediction time is the time required to predict the parameters when the base curve of the eyeglass lens selected based on the wearer's prescription is 8.00.

When the base curve of the eyeglass lens selected based on the wearer's prescription was 8.00, the parameter prediction time using the first parameter prediction model was 250 ms to 1500 ms.

When the base curve of the spectacle lens selected based on the wearer's prescription was 8.00, the parameter prediction time using the second parameter prediction model was 250 ms to 1500 ms.

When the base curve of the eyeglass lens selected based on the wearer's prescription was 8.00, the parameter prediction accuracy using the third parameter prediction model was -2% to +3%.

According to the verification results shown in Figure 6, the prediction time using the second parameter prediction model is shorter than the prediction time using the third parameter prediction model. In other words, it can be seen that the prediction time using the parameter prediction model can be shortened by using a prediction model that learns from data with a narrow learning range that includes the spherical refractive power corresponding to the target base curve.

Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It is clear to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can be included within the technical scope of the present invention.

For example, the above embodiment includes one design server 100 as the design system. However, the design system may include multiple design servers 100.

For example, the above embodiment has been described using the design server 100 as an example of a design system. However, the design system is not limited to the design server 100. The design system may include, for example, multiple servers. In a design system including multiple servers, software modules that function in the design server 100 may function distributed across the multiple servers.

For example, in the design server 100 of the above embodiment, the CPU 110 functions as the parameter prediction unit SM2. However, the CPU 110 does not have to function as the parameter prediction unit SM2. For example, the design server 100 may be equipped with an AI (artificial intelligence) chip. The AI chip is a semiconductor for speeding up AI calculation processing. When the design server 100 is equipped with an AI chip, the AI chip may function as the parameter prediction unit SM2.

For example, in the design server 100 of the above embodiment, the CPU 110 functions as the processing time prediction unit SM3. However, the CPU 110 does not have to function as the processing time prediction unit SM3. For example, the design server 100 may be equipped with an AI chip. If the design server 100 is equipped with an AI chip, the AI chip may function as the processing time prediction unit SM3.

For example, in the design server 100 of the above embodiment, the CPU 110 functions as the change data prediction unit SM4. However, the CPU 110 does not have to function as the change data prediction unit SM4. For example, the design server 100 may be equipped with an AI chip. If the design server 100 is equipped with an AI chip, the AI chip may function as the change data prediction unit SM4.

For example, the design server 100 in the above embodiment executes the process of S106 shown in FIG. 4, and then executes the process of S107. After executing the process of S106, the design server 100 may execute the process of S103 and subsequent steps again.

For example, in the above embodiment, the base curve of the spectacle lens is a value selected based on the wearer's prescription. However, the base curve of the spectacle lens may be a value desired by the wearer.

For example, the data that serves as the teacher model for the parameter prediction model may have a different learning rate within a specific range of values that the data can take and a different learning rate outside the specific range. In the field of optimization, the learning rate is also referred to as the step size or step width. For example, if there is a bias in the popularity of eyeglass frames, the parameter prediction model will increase the learning rate for popular shapes and decrease the learning rate for unpopular shapes for the frame shape data D34, thereby increasing the prediction accuracy.

For example, in the above embodiment, the data group serving as training data for the parameter prediction model may include processing machine data relating to a processing machine that processes eyeglass lenses. By using a parameter prediction model trained on a data group including processing machine data as training data, the parameter prediction model can avoid predicting parameters for shapes that cannot be processed by the processing machine.

For example, reinforcement learning of parameter prediction models works well with simulation. By using a simulator that mimics a real wearer, eyeglass lenses, and eyeglass frames, the parameter prediction model can efficiently perform many trials and observe the situation on the simulator. By using a simulator, the parameter prediction model can select options that would not be possible in the real world, and by actually experiencing failure, it can learn not to select the failing option.

The order of execution of each process, such as operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings, is not specifically stated as "before," "prior to," or the like. It should be noted that the order of execution of each process may be any order, as long as the output of a previous process is not used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is explained using "first," "next," or the like for convenience, this does not mean that it is essential to carry out the process in that order.

100: Design server, 110: CPU, 120: Main memory, 130: Input/output interface, 140: Communication device, 150: Storage, CN: Computer network, D1: Eyeglass lens data, D2: Prescription data, D3: Eyeglass frame data, D4: Parameter data, D5: Shape data, D11: Base curve data, D12: Lens outer diameter data, D13: Center thickness data, D14: Front vertex power data, D21: Spherical power data, D22: Cylindrical power data, D23: Cylindrical axis angle data, D24: Addition power data, D31: Curvature angle data when worn, D32: When worn Forward tilt angle data, D33: vertex distance data, D34: frame shape data, D35: visual point data, D36: interpupillary distance data, OS: optician, OT: optician terminal, PM1: first parameter prediction model, PM2: second parameter prediction model, PM3: third parameter prediction model, PM4: fourth parameter prediction model, PM5: fifth parameter prediction model, PM6: processing time prediction model, PM7: change data prediction model, SM1: data group receiving unit, SM2: parameter prediction unit, SM3: processing time prediction unit, SM4: change data prediction unit, SM5: rear surface shape calculation unit, SM6: shape data transmission unit

Claims (8)

A spectacle lens design system for designing a spectacle lens, comprising:
a parameter prediction unit that predicts parameters for optically calculating the shape of the rear surface of the eyeglass lens from a data group including eyeglass lens data related to the eyeglass lens using a parameter prediction model;
a rear surface shape calculation unit that calculates the shape of the rear surface of the eyeglass lens,
the parameter prediction model is a prediction model machine-learned using the data group and the parameters as training data,
The posterior surface shape calculation unit calculates a shape of a posterior surface of the eyeglass lens using the parameters predicted by the parameter prediction unit. The eyeglass lens design system according to claim 1, wherein the data group includes prescription data relating to a prescription of a wearer of the eyeglass lens. The eyeglass lens design system according to claim 1, wherein the data group includes eyeglass frame data relating to an eyeglass frame desired by a wearer. The eyeglass lens design system according to claim 1, wherein the rear surface shape calculation unit calculates the shape of the rear surface of the eyeglass lens by ray tracing calculation and wavefront calculation. The parameter prediction model is
A first parameter prediction model machine-learned using the data group including the eyeglass lens data having a base curve of the eyeglass lens as a first value as training data;
A second parameter prediction model machine-learned using the data group including the eyeglass lens data having a base curve of the eyeglass lens as a second value as training data,
The parameter prediction unit
When a base curve of a spectacle lens selected based on a prescription of a wearer is the first value, predicting the parameter from the data group using the first parameter prediction model;
2. The eyeglass lens design system according to claim 1, wherein when a base curve of an eyeglass lens selected based on a wearer's prescription is the second value, the parameter is predicted from the data group using the second parameter prediction model. a processing time prediction unit that predicts whether the shape of the rear surface of the eyeglass lens can be determined within a predetermined time from the parameters using a processing time prediction model;
the processing time prediction model is a prediction model machine-learned using the parameters and performance data on whether the shape of the rear surface of the eyeglass lens was determined within a predetermined time as training data,
The eyeglass lens design system according to claim 1 , wherein the parameter prediction unit, when it is predicted that the shape of the rear surface of the eyeglass lens cannot be determined within a predetermined time, changes a portion of the data in the data group and re-predicts the parameters. A data group receiving unit that receives the data group;
a shape data transmission unit that transmits shape data relating to the shape of the rear surface of the eyeglass lens;
The eyeglass lens design system according to claim 6 , wherein the processing time prediction unit predicts whether the shape of the rear surface of the eyeglass lens can be determined within a remaining time of a timeout period from when the data group is received until when the shape data is transmitted. a change data prediction unit that predicts which data among the data group should be changed based on the remaining time by using a change data prediction model;
the change data prediction model is a prediction model machine-learned using the performance data and the data group as training data,
The eyeglass lens design system according to claim 7 , wherein the parameter prediction section re-predicts the parameters by modifying data predicted by the modified data prediction section.

Images