Coffee is a critical agricultural commodity to be used to produce a premium beverage to serve people worldwide. Coffee microbiome turned to be an essential tool to improve the bean quality through the natural fermentation. Therefore, understanding the microbial diversities could create the final product's better quality. This study investigated the natural microbial consortium during the wet process fermentation of coffee onsite in Thailand to characterize the microorganisms involved in correlation toward the biochemical characteristics and metabolic attributes. Roasting is another important step in developing the complex flavor/ aroma that make coffee to be enjoyable. During the roasting process, the beans undergo many complex and alternatively change in the physicochemical properties from the gained substances in the fermentation process. The changing in the formation of the substances responsible for the sensory qualities, physicochemical/ aroma attributes as well as the health benefits of the final product. Using the starter culture could also develop the distinguished characteristics of coffee (Research collaboration with Van Hart company)
-
คณะครุศาสตร์อุตสาหกรรมและเทคโนโลยี
This project presents the development of a single-frequency GPS-based total electron content measurement tool. It applies theories related to total electron content in the ionospheric layer and the measurement of total electron content using GPS time delay to design the single-frequency GPS total electron content measurement tool. The tool consists of an antenna, a single-frequency GPS satellite receiver, a data processing unit for evaluating and calculating total electron content, and a display unit for showing total electron content data. The performance of the single-frequency GPS total electron content measurement tool is tested by comparing it with total electron content data obtained from the International Reference Ionosphere (IRI) model, which is a global reference model for electron content. The tool is also put to practical use. The results of the comparison and practical applications conclude that the single-frequency GPS-based total electron content measurement tool can be effectively utilized, with the difference from the IRI model being 50 TECU
คณะวิทยาศาสตร์
This project presents the development of an automatic recycling machine for plastic bottles and cans, utilizing Machine Learning for packaging classification through image processing, integrated with smart sensor systems for quality inspection and operation control. The system connects to a Web Application for real-time monitoring and control. Once the packaging type is verified, the system automatically calculates the refund value and processes payment through e-wallet or issues cash vouchers. The system can be installed in public spaces to promote waste segregation at source, reduce contamination, and increase recycling efficiency. It also provides financial incentives to encourage public participation in waste management. This project demonstrates the potential of combining Machine Learning and smart sensor systems in developing accurate, convenient, and sustainable waste management solutions.
คณะวิทยาศาสตร์
This special problem aims to study and compare the performance of predicting the air quality index (AQI) using five ensemble machine learning methods: random forest, XGBoost, CatBoost, stacking ensemble of random forest and XGBoost, and stacking ensemble of random forest, SVR, and MLP. The study uses a dataset from the Central Pollution Control Board of India (CPCB), which includes fifteen pollutants and nine meteorological variables collected between January, 2021 and December, 2023. In this study, there were 1,024,920 records. The performance is measured using three methods: root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination. The study found that the random forest and XGBoost stacking ensemble had the best performance measures among the three methods, with the minimum RMSE of 0.1040, the minimum MAE of 0.0675, and the maximum of 0.8128. SHAP-based model interpretation method for five machine learning methods. All methods reached the same conclusion: the two variables that most significantly impacted the global prediction were PM2.5 and PM10, respectively.