Inspection apparatus, unit selection apparatus, inspection method, and computer-readable storage medium storing an inspection program

An inspection method implementable with a computer, the method comprising:

• obtaining a target image of a target product to be inspected;
• extracting an attention area from the obtained target image using a first estimation model, the first estimation model being generated by training to extract, as the attention area, an image area having a probability of a defect being comprised therein, based on first training data, the first training data comprising a plurality of first training images of defect-free products obtained in a target inspection environment;
• performing a computational process with a second estimation model using the attention area extracted from the target image by the first estimation model,
• the second estimation model being generated by training to determine a likelihood of the defect being comprised therein based on second training data comprising a plurality of second training images of defects,
• the computational process with the second estimation model comprising a plurality of projection processes performed on the target image, the plurality of projection processes including comprising projecting the target image into different spaces with lower dimensions to generate a plurality of feature maps with different dimensions,
• the extracted attention area being integrated into at least one of the plurality of feature maps in the computational process with the second estimation model;
• determining whether the target product has a defect based on a computation result from the second estimation model;
• and
• outputting information indicating a result of determining whether the target product has a defect.

A non-transitory computer-readable storage medium storing an inspection program, which when read and executed, causes a computer to perform comprising:

• obtaining a target image of a target product to be inspected;
• extracting an attention area from the obtained target image using a first estimation model, the first estimation model being generated by training to extract, as the attention area, an image area having a probability of a defect being comprised therein, based on first training data, the first training data comprising a plurality of first training images of defect-free products obtained in a target inspection environment;
• performing a computational process with a second estimation model using the attention area extracted from the target image by the first estimation model,
• the second estimation model being generated by training to determine a likelihood of the defect being comprised therein, based on second training data comprising a plurality of second training images of defects,
• the computational process with the second estimation model comprising a plurality of projection processes performed on the target image, the plurality of projection processes comprising projecting the target image into different spaces with lower dimensions to generate a plurality of feature maps with different dimensions,
• the extracted attention area being integrated into at least one of the plurality of feature maps in the computational process with the second estimation model;
• determining whether the target product has a defect based on a computation result from the second estimation model;
• and
• outputting information indicating a result of determining whether the target product has a defect.