Deep learning technology in multimodal remote sensing data fusion to analyze crop growth and economic forecast

Abstract

This paper integrated multimodal remote sensing (RS) data with deep learning to develop a maize growth analysis and income prediction model based on CNN (Convolutional Neural Network)-Attention and Bi-LSTM (Bidirectional Long Short-Term Memory). Utilsing Landsat 8, Sentinel-2, and Sentinel-1 satellite data, the CNN-Attention network extracts vegetation features such as NDVI (Normalised Difference Vegetation Index), EVI (Enhanced Vegetation Index), and canopy density for corn growth stage identification and prediction. The study combined these features with meteorological, soil, and market data and fed them into a Bi-LSTM model for time series analysis to forecast corn income. The method achieved 98.3% accuracy in growth stage classification, with an RMSE (Root Mean Squared Error) of 0.04 and R² of 0.94 for canopy coverage prediction under 10-fold cross-validation, and an RMSE of 0.13 and R² of 0.94 for income prediction, showing high stability over time. Overall, this research supports data-driven precision agriculture through real-time monitoring and reliable economic forecasting.

Plant phenotyping relevance

衛星リモートセンシングと深層学習による作物生育段階・キャノピー密度の推定手法が中心的に開発・検証されており、植物状態の定量化を含むため。収益予測も扱うが、植物フェノタイプ推定部分が明示されている。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.