A Simple and Efficient Deep Learning Architecture for Corn Yield Prediction

Agricultural yield estimation plays a very important role in providing the necessary planning and organization to feed the growing world population. Although past to present regression-based, simulation-based and hybrid methods are proposed for agricultural yield estimation, Machine Learning (ML) methods have been used in recent years. Among the ML techniques, Deep Learning (DL) is one of the most prominent method recently. In DL techniques, either satellite data on which data reduction methods are applied, as well as weather, soil, etc. data or directly processed large satellite images are used. The aim of this study is to propose an efficient and fast DL architecture for corn yield estimation. In the study, only histogram-based satellite data and yield information were used. The proposed architecture combines the Convolutional Neural Network (CNN)’s ability to learn depth dependent features and detect different feature sets using different kernel sizes in parallel. Proposed architecture is experimented for corn yield estimation of 13 states in the U.S. Corn Belt at county-level. Corn yield from 2018 to 2021 are estimated using only Moderate Resolution Imaging Spectroradiometer’s (MODIS) last ten years Surface Reflectance (SR) and Land Surface Temperature (LST) data and last ten years yield data obtained from United States Department of Agriculture (USDA) Quick Stats. The preliminary results show that the proposed architecture uses less data and works faster, giving results close to or even better than the studies in the literature.