情報処理学会 第87回全国大会

6U-08
A Study On Improving Similar Issue Of Inter-classes In Food Ingredients Recognition
○趙  毅(岩手県大)
This study focuses on improving the classification of food ingredients with high inter-class similarity using machine learning techniques. By analyzing confusion matrices, grouping similar ingredients, and training specialized classification models, the proposed system enhances recognition accuracy. EfficientNetB0 and MobileNetV2 architectures were optimized for reclassification tasks. Evaluation on the SI110 dataset demonstrated significant improvements in identifying challenging ingredients, offering a robust approach for accurate food recognition.