食肉処理ロボットシステムにおける認識技術の開発

―第２報：深層学習による豚枝肉分割位置の検出―

抄録

Automation of the meat processing is highly desired due to dangers of using sharp cutting tools and involves heavy labor. In this study, we aim to implement the highly accurate method for cut point detection by integrating deep learning in the recognition unit of a meat processing system. The newly developed key point detection method which infers feature point from surrounding patterns was used for detecting cut point in the prime cut process. First, we conducted the preliminary examination of key point detection using existing datasets. However, the detection accuracy was low due to many errors in the teaching points, so we optimized the imaging system and teaching points. As a result, the accuracy of detection was improved to 97.2% from 63.6% at maximum.