Yield prediction of field vegetables using functional regression model and deep learning algorithm

Modern agriculture requires the development of new AI-driven farming methods that can achieve maximum yields while minimizing the use of fertilizers, energy resources, and labor. Therefore, this study aims to propose smart agricultural technologies that can optimize economic efficiency by accurately predicting crop yields in advance through frequent monitoring of the growth status of field vegetables, such as onions and garlic, and appropriately adjusting the supply and demand of the harvested products. An experimental unit was set up at the Research Institute in Muan-gun, Jeollanam-do. The fertilizer levels were set at four different levels based on a standard amount. Multispectral images were collected at regular intervals during the cultivation period of onions and garlic in each plot, and 10 vegetation indices were extracted from these images. As a result of the experiment, for onions, the model combining 1D-CNN with the sheath length variable showed the best performance, followed by the model combining 1D-CNN with RECI and the model combining 1D-CNN+BiLSTM with MNGRD. For garlic, the model combining 1D-CNN with the sheath length variable was also the most accurate, followed by the model combining 1D-CNN with NGRD and the model combining 1D-CNN+BiLSTM with the sheath length variable.