Quantification of favorableness to foods by information processing on human sense

对啤酒、银杏清酒、咖啡等食品的喜好程度进行了量化，得到了以下结果：(1)建立了咖啡的品质模型。收集了67个滴制样品的分析数据和感官评定数据。将模糊神经网络(FNN)应用于建模时，得到的FNN模型与传统的多元回归分析模型相比，具有更高的感官评价精度。(2)将FNN和HFNN(递阶模糊神经网络)应用于不同银杏清酒样品的感官评价。FNN和HFNN模型的估计误差分别约为10%和7%。用模糊神经网络和高频神经网络选择的输入变量与专家的经验吻合较好。通过模糊规则分析，这些输入变量的定性效果与专家经验基本一致(…3)利用模糊神经网络建立了啤酒和啤酒酿造过程的感官评价模型，并将遗传算法和扫描算子优化模型的新方法与传统的参数增量法进行了比较。结果表明，该方法简化了模型结构，提高了模糊规则的可靠性，加快了建立高精度模型的计算速度(约为常规方法的10倍)，有利于模型的优选。选取重要变量作为输入变量，得到的建模模糊规则与过程操作员获取的知识库吻合较好。(4)开展了味觉传感器的基础性研究。克隆了味觉受体基因，建立了重组大肠杆菌味觉受体蛋白的纯化方法。将纯化的蛋白质浸泡在传感器尖端(I.D.3 mm)的表面等离子体共振分析仪，可以检测到分析仪输出的变化。因此，与味觉受体这一特定感官分子相关联的味觉传感器有可能被开发出来。较少

Quantification of favorableness to foods such as beer, Ginjo sake and coffee, was carried out and the following results were obtained.(1) Quality modeling of coffee was constructed. The analytical data from 67 samples obtained by dripping and those sensory evaluation data were collected. When fuzzy neural network (FNN) was applied to the modeling, the FNN model acquired showed the higher accuracy on estimation of sensory evaluation, compared with the model by the conventional method, multi-regression analysis.(2) FNN and HFNN (hierarchical fuzzy neural network) were applied in order to construct the models estimated from the analysis data for the sensory evaluations of various Ginjo sake samples. Errors estimated by FNN and HFNN models were about 10% and 7%, respectively. Selected input variables using FNN and HFNN were in good agreement with expert's experiences. By the analysis of fuzzy rules, qualitative effects of these input variables were almost the same as expert's experiences.( … More 3) Models for sensory evaluation of beer and beer brewing process were constructed using FNN.A new method for optimal model selection using genetic algorithm and SWEEP operator method was compared with a conventional method using parameter increasing method. As the result, the new method was useful for the optimal model selection by simplifying the model structure, improving the reliability of fuzzy rules and fastening the calculation speed (about 10 times as fast as conventional method) for constructing the model with high accuracy. The important variables were selected as the input variables and the obtained fuzzy rules in modeling coincided well with knowledge data bases acquired by process operators.(4) Fundamental research on tasting sensor was carried out. The gene encoding a gustatory receptor was cloned and the purification method of the receptor protein from recombinant Escherichia coli was established. When the purified protein was dipped on the sensor tip (i.d. 3mm) of surface plasmon resonance analyzer, the change of analyzer output could be detected. Therefore, the tasting sensor associated with a specific sensing molecule, gustatory receptor, is possible to be developed. Less