Machine Learning Based Classification Models in Smart Farming ; スマート農業における機械学習に基づく分類モデル

九州工業大学 ; 博士（情報工学） ; 1 Introduction|2 Machine Learning|3 Smart Farming|4 Farming Product Quality Inspection by Object Group Classification|5 Farmer Activity Monitoring by Temporal Data Classification|6 Conclusion, Limitations and Future Works ; The development of information systems and technology, referred to as Machine Learning, has numerous applications across various industries. One application is within Smart Farming, an agricultural concept based on precision agriculture. The application utilizes platforms connected to technological automation devices. Big data management, Machine Learning, Artificial Intelligence (AI), and the Internet of Things (IoT) facilitate data processing for optimizing the quality and quantity of production to maximize farming methods, agriculture technologies, and human resources. Data obtained from farming locations will be beneficial if presented in the proper format at the appropriate time. The direction for Machine Learning (ML) implementation techniques that operate on simple IoT devices and accomplish simple processing on-site, which enable information to flow to and from remote and agricultural farming sites, has become a fundamental requirement. This dissertation proposes two approaches to implementing Machine Learning (ML) techniques within IoT communication schemes for practical, increased precision, solving real-world problems, improving farming operating efficiency, and providing robust solutions. With consideration of energy-efficient, the number of the evaluated dataset, time calculation, and computational performance. The first approach is implementing a feature classification framework focusing on the isolated groups of objects for agricultural products. On digital images of cocoa beans as a farmed product, we have demonstrated a method of textural feature research. In terms of feature extraction, the method contrasts the Gray Level Co-occurrence Matrix (GLCM) cooccurrence matrix features with the Convolutional Neural Network (CNN) method. Moreover, we used a series ...