Mining Spatio-Temporal Knowledge of Climate for Dendrobium Officinale in Greenhouse Cultivation

摘要

Background: This paper presents a novel spatio-temporal learning method to generate knowledge of fine-grained cultivation of Dendrobium officinale from large climatic sensor data. In recent years, datadriven precision agriculture, which is strongly connected with data mining techniques, has a great impact on traditional farming. Most of the environmental control methods in the green house are based on humanexperience. Methods: In the paper, fine-grained rules have been proposed to provide precise climate control in temporal dimensions during the growth of plants. Spatial and temporal information were combined together and analyzed. An unsupervised merging algorithm on time segments was proposed to learn optimal tags for classification, and then new data samples consisting of clustering labels were prepared for knowledge mining from the raw sensor data. Finally the climatic rules were built in the form of decision trees. Results: The experimental results show that about 80% climatic conditions in the past successful cultivation can be replicated to guide future cultivation by our decision rules. Conclusion: The knowledge cultivation can guarantee similar quality of Dendrobium officinale since it is especially important for the production of herb medicine.