A deep convolutional neural network(Deep CNN)was developed for evaluating the quality of rice for sake. The surface and inside images of the grains were classified into four(whole, immature, damaged and broken grains)and two(non-cracked and cracked grains)classes, respectively, by a skilled analyst. The architecture of Deep CNN was based on the connection of eight convolutional layers, four max pooling layers, and two fully connected layers, and the appropriate number of layers and dropout rate were verified by hyper-parameter tuning. By the application of the optimized model to classify the evaluation data, accuracy of the surface and inside image classification was rated 89.9% and 86.1%, respectively. In addition, the features of Deep CNN focused on in the image were confirmed by using Grad-CAM, and it was shown Depp CNN had the similar criteria for classification to the criteria of visual judgments. This result clarified that Deep CNN is useful for evaluating the quality of rice grains for sake as an alternative to existing methods.
Sequential earlier-period-rice analysis data gathered from the 2002–2019 crop years, comprised of 707 samples in total, were examined focusing on a comparison within sake rice varieties. We discovered varied rice properties in Japanese sake brewing among different rice varieties. The weight of 1,000 rice grains was heavier across Hanafubuki, Okuhomare, and Hyogokitanishiki, followed by Yukimegami, Homarefuji, Akitasakekomachi, Yamadahishiki, Kitashizuku, Senbon-nishiki, and Hattan-nishiki. Water absorption during 20 minute periods was higher in Kyonohana and Dewanosato, followed by Homarefuji, Gin-noyume, Hyogokitanishiki, Hattan-nishiki, Yamadahishiki, Ipponjime, and Hanafubuki. Brix was particularly high in Dewanosato, followed by Kyonohana, Homarefuji, Akitasakekomachi. The ratio of Brix/folmol nitrogen was higher in Dewanosato, followed by Yukimegami, Yamadanishiki, Gin-noyume, Nakateshinsenbon, and Homarefuji. Crude protein was higher in Okuhomare, Ginpu, Kitashizuku, Fukuhikari, and Hyogokitanishiki, but lower in Yukimegami, Dewanosato, Yamadanishiki, Nakateshinsenbon, Akitasakekomachi, Gin-noyume, and Sanukiyoimai. Brewing properties across all rice varieties analyzed did not considerably vary from the properties of rice harvested within the same production area, indicating that sake rice variety strongly affects rice properties upon sake brewing.
In the present study, we investigated the annual variation in the brewing properties of sake rice utilizing earlier-period-rice analysis data gathered from 2002–2019 across all rice varieties analyzed. The annual variation for water absorption during a 20 minute period of table rice was higher than sake rice. Water absorption during both 20 minute and 120 minute periods of a large amount of rice varieties was higher among rice harvested during the year 2010, when the highest mean August temperature was recorded in Japan. The starch digestibility(Brix) of a large amount of rice varieties was lower in the years 2007, 2010, 2012, and 2019, during which higher temperatures from July to September contrasted with other crop years. Brix values of Gohyakumangoku varied widely within a year depending upon the production district, indicating that the variety is likely to be strongly affected by meteorological conditions in regards to its digestibility. The annual variation of crude protein and potassium concentrations within white rice varied depending on rice variety and the production district. Brix values of Yamadanishiki and Gohyakumangoku from 2008 to 2018 were strongly and negatively correlated with the mean data of sake cake ratios for junmai ginjo and ginjo of the corresponding year, suggesting that the Brix value would be available for the estimation and improvement of premium sake brewing practices. Taken altogether, by analyzing the annual variation, we discovered that the brewing properties of sake rice can be considerably altered by meteorological conditions, and that the effects of meteorological conditions upon brewing properties vary depending on the type of rice variety in question.