A crop model based on dual attention mechanism for large area adaptive yield prediction

Accurate estimation of crop yield over a large geographical area is crucial for food security and sustainable development. With the rapid advancement of machine learning, methods for agricultural yield prediction based on deep learning models have become a primary means of improving prediction accur...

Full description

Saved in:
Bibliographic Details
Main Authors: Wei Xiang, Long Long, Zichen Liu, Feng Dai, Yucheng Zhang, Hu Li, Lin Cheng
Format: Article
Language:English
Published: Elsevier 2025-08-01
Series:Smart Agricultural Technology
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S277237552500190X
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurate estimation of crop yield over a large geographical area is crucial for food security and sustainable development. With the rapid advancement of machine learning, methods for agricultural yield prediction based on deep learning models have become a primary means of improving prediction accuracy. Although existing models have improved accuracy by increasing model complexity and coupling different deep learning models, their generalization performance is poor due to significant spatial differences in crop growth environments, making it difficult to explore common features of crop environments in different regions.To address this issue, this paper comprehensively considers crop growth cycles and environmental factors such as soil and weather, presenting a large-scale crop yield prediction model based on an attention mechanism.The model consists of two modules: time attention module and feature attention module. The time attention module extracts temporal features of crop growth in different regions, while the feature attention module extracts environmental factor features in different regions. Together, the two modules extract spatial differences in different regions. Next, we will integrate the features extracted from these two blocks and feed them into a multi-layer perceptron for yield prediction. We conducted experiments on soybean production and environmental data from 1045 soybean counties across 12 states in the United States from 1980 to 2018, covering 39 years. The experimental results demonstrate that the model proposed in this paper exhibits significantly lower root mean square error (RMSE) and higher correlation coefficients compared to other models when predicting crop yields for the years 2016, 2017, and 2018.
ISSN:2772-3755