Training data generation device and training data generation program
A training data generation device generates training data usable in machine learning. A learned model using the training data generated by the training data generation device is used in an inspection device for determining whether an inspection target is a normal product by inputting an image capturing the inspection target into the learned model. The training data generation device includes: a determination-target image extraction unit that extracts, from an input image, one or more determination-target images containing a determination target that satisfies a predetermined condition; a sorting unit that associates, on the basis of sorting the inspection target captured in the determination-target image, each of the determination-target images and a result of the sorting with each other; and a training data memory unit that stores training data in which each of the determination-target images and a result of the sorting are associated with each other.
1. A training data generation device generating training data to be used in machine learning, a learned model generated by the machine learning using the training data being used in an inspection device that performs inspection for determining whether or not an inspection target is a normal product by inputting an image capturing the inspection target into the learned model, the training data generation device comprising:
a feed unit configured to sequentially feed a plurality of inspection targets, including the inspection target, through an inspection area;
a detector configured to detect a spatial distribution of the electromagnetic waves that reach the detector through the inspection area, the detector including either a line sensor or a Time Delay Integration (TDI) sensor;
an image construction unit configured to construct electromagnetic-wave intensity contrast distribution as an image of the inspection target fed by the feed unit, while the feed unit sequentially feeds the plurality of inspection targets;
a determination-target image extraction unit configured to extract, from an input image capturing the inspection target, one or more determination-target images containing a determination target that satisfies a predetermined condition, using the image of the inspection target constructed by the image construction unit as the input image;
a sorting unit configured to associate, on the basis of a sorting operation of sorting the inspection target captured in the determination-target image into either a normal product or a non-normal product, each of the determination-target images and a result of the sorting with each other; and
a training data memory unit configured to store training data in which each of the determination-target images and a result of the sorting are associated with each other, wherein
the determination-target image extraction unit comprises:
an image processing unit configured to perform image processing on the input image and outputs a processed image;
a determination-target identification unit configured to identify the determination target contained in the processed image on the basis of a predetermined determination condition; and
an image cutting-out unit configured to cut out and output, from the processed image, the determination-target image containing the determination target identified by the determination-target identification unit, and
the image cutting-out unit is configured to
identify a minimum rectangle containing the determination target identified by the determination-target identification unit, and
output, as the determination-target image, an image obtained by scaling an image contained in the minimum rectangle to a predetermined size.
2. The training data generation device according to claim 1 , wherein
the determination-target image extraction unit is configured to
normalize the brightness of the determination-target image and
output the determination-target image.
3. The training data generation device according to claim 1 , wherein
the training data generation device further comprises:
a display; and
a display control unit configured to control display performed by the display, and wherein
the display control unit is configured to
cause a first area of the display to display a plurality of the determination-target images in sequence and
cause a second area of the display to display the input image with an indication of an area corresponding to the determination-target image being superimposed on the input image.
4. The training data generation device according to claim 3 , wherein,
in response to one of the plurality of determination-target images displayed in the first area of the display being selected, the display control unit is configured to cause an indication of an area corresponding to the selected determination-target image to be displayed in the input image displayed in the second area of the display.
5. A non-transitory computer-readable storage medium storing a computer program for causing a computer executing the computer program to be configured as a training data generation device for generating training data to be used in machine learning, a learned model generated by the machine learning using the training data being used in an inspection device that performs inspection for determining whether or not an inspection target is a normal product by inputting an image capturing the inspection target into the learned model, the training data generation device comprising:
a feed unit configured to sequentially feed a plurality of inspection targets, including the inspection target, through an inspection area;
a detector configured to detect a spatial distribution of the electromagnetic waves that reach the detector through the inspection area, the detector including either a line sensor or a Time Delay Integration (TDI) sensor;
an image construction unit configured to construct electromagnetic-wave intensity contrast distribution as an image of the inspection target fed by the feed unit, while the feed unit sequentially feeds the plurality of inspection targets;
a determination-target image extraction unit configured to extract, from an input image capturing the inspection target, one or more determination-target images containing a determination target that satisfies a predetermined condition, using the image of the inspection target constructed by the image construction unit as the input image;
a sorting unit configured to associate, on the basis of a sorting operation of sorting the inspection target captured in the determination-target image into either a normal product or a non-normal product, each of the determination-target images and a result of the sorting with each other; and
a training data memory unit configured to store training data in which each of the determination-target images and a result of the sorting are associated with each other, wherein
the determination-target image extraction unit comprises:
an image processing unit configured to perform image processing on the input image and outputs a processed image;
a determination-target identification unit configured to identify the determination target contained in the processed image on the basis of a predetermined determination condition; and
an image cutting-out unit configured to cut out and output, from the processed image, the determination-target image containing the determination target identified by the determination-target identification unit, and
the image cutting-out unit is configured to
identify a minimum rectangle containing the determination target identified by the determination-target identification unit, and
output, as the determination-target image, an image obtained by scaling an image contained in the minimum rectangle to a predetermined size.