INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, INFORMATION PROCESSING PROGRAM, LEARNING DEVICE, LEARNING METHOD AND LEARNED MODEL

To solve such a problem that many images are required when performing machine learning using an image of a tablet having a defect.SOLUTION: A learning device comprises an image restoration unit which performs learning with deep learning so as to be able to generate a restoration image in which a concealed portion is restored from image data in which a portion of each of a plurality of learning images is concealed obtained by imaging a normal tablet that has a secant line and has no defect like a scratch from the oblique direction. The secant line passes through a center of a secant line surface on which the secant line is formed in the tablet and extends straight to both ends of the secant line surface, and a position of the secant line can be recognized from an end of the secant line appearing on a side surface of the tablet. With this, the device can determine whether the tablet is normal or abnormal by comparing the restoration image of the inspection image obtained by imaging the tablet that is unclear to be normal or abnormal from the oblique direction with the inspection image. Also, the device can detect a defect existing on the side surface of the tablet, and can perform learning and inspection while recognizing the position of the secant line even when the secant line surface faces the rear side.SELECTED DRAWING: Figure 10 【課題】欠陥を有する錠剤の画像を用いて機械学習を行う場合、多数の画像が必要になる。【解決手段】この学習装置は、割線を有する傷等の欠陥の無い正常な錠剤が斜め方向から撮像された複数の学習画像のそれぞれの一部が隠された画像データから、隠された一部が復元された復元画像を生成できるように、ディープラーニングにより学習する画像復元部を有する。また、割線は、錠剤のうちの割線が形成される割線面の中心を通り、割線面の両端まで真っ直ぐに延び、かつ、錠剤の側面に現れる割線の端部から、割線の位置を認識できる。これにより、正常か異常か不明な錠剤が斜め方向から撮像された検査画像の上記の復元画像を検査画像と比較することによって、錠剤が正常か異常かを判定できる。また、錠剤の側面に存在する欠陥をも検出でき、さらに、割線面が裏側を向いている場合でも、割線の位置を認識しつつ、学習および検査を行うことができる。【選択図】図１０