[2Win5] Poster session 2

[2Win5-70] Multimodal learning of food compression and 4D chewing measurements toward sophistication of food texture analysis based on machine learning

〇Makoto Takemasa¹, Chang Syousei¹ (1.Tokyo Denki University)

Keywords:Food texture, Deep learning, Multimodal learning, 3D scanner, Mastication

More than 60% of palatability originates from physical properties of foods, food texture. However, it is quite difficult to analyze due to too much complexity and large variation preventing from simple multivariate analysis. Sensory test is not stable enough, and small difference of food texture cannot be detected. In addition, for polymers controlling almost all the food texture, theory and even simulation cannot be applied, especially in thick systems due to strong non-linearity. We measured food textures to build big data. We have shown that deep learning can distinguish small texture differences even. In this study, we tried to make multimodal learning structure to combine different types of texture results, 1-axis compression test, and 4D face shape during mastication, and succeeded to obtain higher classification ability based on multimodal learning.