이중 분기 LSTM 모델 기반 양파 생육 특성 예측

초록

Recent climate change has increased the uncertainty associated with experience-based farming, highlighting the need for accurate prediction of crop growth traits. In this study, we propose a deep learning model that predicts onion growth traits using multi-source agricultural data. The proposed model adopts a dual-branch architecture composed of a dynamic branch based on an LSTM network that processes time-series environmental and crop growth data, and a static branch that incorporates fixed farm-level characteristics such as cultivar type and planting density. To validate the model, we used a dataset collected from three onion farms in Jeonbuk Province during 2024 and 2025. Experimental results show that the proposed model achieved average coefficients of determination() of 0.91 and 0.90 for bulb diameter and fresh bulb weight, respectively, demonstrating strong explanatory power for both traits. These findings confirm the effectiveness of the proposed multi-source data modeling approach and indicate its potential to support yield estimation and decision-making in crop management in precision agriculture

키워드