An image of an inspection surface of a product shot with a first imaging unit is divided into partial area images. For each partial area, a narrow-angle partial area image shot with a second imaging unit is acquired under shooting conditions in a pre-learning table in which the conditions and IDs are set. For each partial area, the narrow-angle partial area image suitable for inspection is selected, an inspection image is generated, and an anomaly of the inspection surface is detected based on the inspection image. For each partial area, whether each condition is effective is determined based on a frequency of the ID included by the inspection image. For each partial area, whether a predetermined condition is established is determined, and, with respect to the partial area in which the predetermined condition is established, a post-learning table is established in which only the condition determined to be effective is set.