S06 - Session O1 - Spatio-temporal characterization of crop growth with multi-category data based on deep learning.

Authors: Alvaro Fuentes *, Sook Yoon, Jongbin Park, Jaesu Lee, Mun Haeng Lee, Dong Sun Park

Monitoring crop growth is critical for sustainable agriculture. Traditionally, this complex analysis has been done manually as a trial and error or by growers' perception. Crop growth is controlled mainly by measuring variables such as temperature, humidity, light, CO 2 , and nutrient availability. Light, temperature and CO 2 are factors that affect photosynthesis. Humidity affects the opening of the stomata and thus uptake of CO 2 . There have been several attempts to model this complex relationship but without any successful evidence on the results. Also, there are various challenges for creating the optimal conditions for crop growth. Therefore, encouraged by the recent advances in precision agriculture and deep learning, we seek further improvements to facilitate the efficient use of resources and avoid losses caused by internal or external conditions that affect the growth process. Our proposed work introduces a learnable system based on deep learning technology to understand these factors' influence and automatically determine the appropriate conditions of crop growth in the spatio-temporal domain using multi-category data. To demonstrate the performance of our research, we collected data from various sensors installed in controlled tomato greenhouse environments in South Korea. Our model enables crop growth prediction based on the measured variables and formalizes a systematic and learnable way to understand the growing conditions of crops by combining spatio-temporal characterization of multi-category data with existing deep learning-based techniques. Our research can efficiently help farmers and researchers understand changes in plant behavior under certain conditions towards achieving maximum productivity while avoiding losses caused during the growth process.