Taste Bud-Inspired Single-Drop Multitaste Sensing for Comprehensive Flavor Analysis with Deep Learning Algorithms.

Area of Science:

• * Sensor technology
• * Artificial intelligence
• * Food science

Background:

• * Electronic tongue (E-tongue) systems aim to mimic human taste perception but face challenges in evaluating complex taste interactions and ensuring reliability in dynamic conditions with error-prone data.
• * Accurately assessing taste is complicated by synergistic and suppressive effects between taste compounds.
• * Dynamic conditions and high deviation error data pose significant hurdles for current E-tongue technologies.

Purpose of the Study:

• * To develop a bioinspired artificial electronic tongue (E-tongue) system that overcomes existing limitations in taste evaluation and reliability.
• * To integrate multiple taste sensor arrays and a deep-learning algorithm for sophisticated taste interpretation.
• * To enhance the accuracy and robustness of E-tongue systems in real-world applications.

Main Methods:

• * A bioinspired artificial E-tongue system was created, emulating human gustatory system with multiple taste sensor arrays.
• * A customized deep-learning algorithm was employed for taste interpretation and analysis.
• * A prototype-based classifier with soft voting was implemented to address disparities and improve classification accuracy.

Main Results:

• * The developed E-tongue system accurately detected four distinct tastes (saltiness, sourness, astringency, sweetness) with good reversibility and selectivity.
• * Taste profiles of six different wines showed similarities to online user reviews, validating the system's taste trend analysis.
• * High classification accuracy was achieved: ~95% for distinguishing wines and ~90% even with over a third of data containing errors.

Conclusions:

• * The bioinspired artificial E-tongue, coupled with deep learning, offers a robust and accurate solution for taste analysis.
• * The system demonstrates significant potential for applications requiring reliable taste detection and classification, even in challenging environments.
• * A deep-learning-powered recommendation system was successfully demonstrated, enhancing user experience through personalized taste insights.