Learning data generation device and defect identification system
A learning data generation device that can generate learning data suitable for learning of an identification model. The learning data generation device has a function of cutting out part of first image data as second image data, a function of generating a two-dimensional graphic corresponding to the area of the second image data and representing a pseudo defect, a function of generating third image data by combining the second image data and the two-dimensional graphic, and a function of assigning a label corresponding to the two-dimensional graphic to the third image data. By using the third image data for learning of the identification model, a highly accurate identification model can be generated.
1. A learning data generation device comprising:
a memory unit; and
a processing unit,
wherein the memory unit is configured to store first image data,
wherein the processing unit is configured to cut out part of the first image data obtained by capturing only an area with a normal pattern as second image data,
wherein the processing unit is configured to generate a two-dimensional graphic corresponding to an area of the second image data and representing a pseudo defect,
wherein the processing unit is configured to generate third image data by combining the second image data and the two-dimensional graphic,
wherein the processing unit is configured to assign a label corresponding to the two-dimensional graphic to the third image data,
wherein the two-dimensional graphic is a first two-dimensional graphic or a second two-dimensional graphic,
wherein the first two-dimensional graphic is generated by specifying shape and color,
wherein the second two-dimensional graphic is generated by cutting out the second image data,
wherein the second two-dimensional graphic is a second polygon, and
wherein the processing unit is configured to assign a third label to the third image data generated by combining the second image data and the second polygon.
2. The learning data generation device according to claim 1 ,
wherein the first two-dimensional graphic is a first polygon, an ellipse, or a double ellipse,
wherein the processing unit is configure to assign a first label to the third image data generated by combining the second image data and the first polygon or the ellipse, and
wherein the processing unit is configure to assign a second label to the third image data generated by combining the second image data and the double ellipse.
3. The learning data generation device according to claim 2 ,
wherein the first polygon comprises a first vertex to an n-th vertex,
wherein n is an integer of 3 or more and 8 or less,
wherein a length of a line segment connecting a point in the first polygon and each of the first vertex to the n-th vertex is a length following a normal distribution,
wherein a mean of the normal distribution is 0.05 times or more and 0.25 times or less a length of a long side of the second image data and a standard deviation of the normal distribution is 0.2 times the mean,
wherein each of R, G, and B of the first polygon is 0 or more and 20 or less (decimal notation) when represented by 256 shades of gray, and
wherein the color of the first polygon has a transmittance of 0% or more and 10% or less.
4. The learning data generation device according to claim 2 ,
wherein a major diameter of the ellipse is 0.05 times or more and 0.25 times or less the length of the long side of the second image data,
wherein a minor diameter of the ellipse is 0.6 times or more and 1.0 times or less the major diameter of the ellipse,
wherein each of R, G, and B of the ellipse is 0 or more and 10 or less (decimal notation) when represented by 256 shades of gray, and
wherein the color of the ellipse has a transmittance of 0% or more and 10% or less.
5. The learning data generation device according to claim 2 ,
wherein a major diameter of the double ellipse is 0.05 times or more and 0.25 times or less the length of the long side of the second image data,
wherein a minor diameter of the double ellipse is 0.6 times or more and 1.0 times or less the major diameter of the double ellipse,
wherein a difference between an outer diameter and an inner diameter of the double ellipse is 5 pixels or more and 15 pixels or less,
wherein R of the double ellipse is 150 or more and 170 or less (decimal notation), G is 60 or more and 80 or less (decimal notation), and B is 20 or more and 40 or less (decimal notation) when represented by 256 shades of gray, and
wherein the color of the double ellipse has a transmittance of 50% or more and 75% or less.
6. The learning data generation device according to claim 1 , wherein the second polygon is a quadrangle cut out from the second image data rotated around a point positioned in the second image data at an angle of 30° or more and 150° or less,
and wherein a center of gravity of the quadrangle is the point and each of a long side and a short side of the quadrangle is 0.1 times or more and 0.25 times or less the length of the long side of the second image data.
7. The learning data generation device according to claim 1 , wherein the processing unit is configured to perform gamma conversion on the third image data; and
wherein the processing unit is configured to perform noise addition or blurring processing on the third image data.
8. A defect identification system comprising:
the learning data generation device;
a database; and
an identification device,
wherein the learning data generation device is configured to cut out part of the first image data obtained by capturing only an area with a normal pattern as second image data,
wherein the learning data generation device is configured to generate a two-dimensional graphic corresponding to an area of the second image data and representing a pseudo defect,
wherein the learning data generation device is configured to generate third image data by combining the second image data and the two-dimensional graphic,
wherein the learning data generation device is configured to assign a label corresponding to the two-dimensional graphic to the third image data,
wherein the first image data, fourth labeled image data, and fifth unlabeled image data are stored in the database,
wherein the identification device is configured to identify a defect contained in the fifth image data on a basis of a learned model,
wherein the two-dimensional graphic is a first two-dimensional graphic or a second two-dimensional graphic,
wherein the first two-dimensional graphic is generated by specifying shape and color,
wherein the second two-dimensional graphic is generated by cutting out the second image data,
wherein the second two-dimensional graphic is a second polygon, and
wherein the processing unit is configured to assign a third label to the third image data generated by combining the second image data and the second polygon.
9. The defect identification system according to claim 8 ,
wherein the learned model is generated on a basis of a learning data set comprising the third image data and the fourth image data.