2T-03
Quality Inspection of Processed Apple Based on Ultraviolet Imaging
○ジハード ラフマワン,オキディッキ プリマ(岩手県大)
Processed apples can be sliced, diced, canned, and pureed. To prepare the apples for these processes pre-processing is required, such as peeling and coring. Most of these operations are performed by machines. However, visual inspection is still necessary to eliminate areas where pre-processing is not sufficient. In this study, we develop an automated inspection based on ultraviolet imaging. Areas with imperfections, such as remaining skin or core, absorb ultraviolet light and appear dark. About 2,000 sliced apples with these imperfections were annotated and a semantic segmentation model was generated based on DeepLabv3. Real-time inspections using this model yielded an average accuracy of 91% in three classes: skin remains, core remains, and bruising damage.