CROP CANOPY SPECTRAL INFORMATION PREDICTION BASED ON EEMD AND CNN-LSTM

The utilization of the Normalized Difference Vegetation Index (NDVI) derived from canopy spectral information during crop growth helps identify growth conditions in farmland areas and facilitates management zone establishment.Targeted fertilizer application is possible based on observed growth patterns.Accurate NDVI prediction from canopy spectral information is crucial for real-time zone delineation and precision fertilization.Long Short-Term Memory (LSTM) neural networks can predict NDVI time series, but have limitations in handling non-stationary signals.A prediction model integrating modal decomposition with Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) is proposed to address these issues.This model uses Ensemble Empirical Mode Decomposition (EEMD) to decompose NDVI data into Intrinsic Mode Functions (IMFs), Convolutional Neural Network (CNN) to extract spatial features from IMFs, and LSTM to predict NDVI.Evaluation against standalone CNN, CNN-LSTM, and EEMD-CNN models using MAPE, RMSE, and R 2 metrics showed the proposed CNN-LSTM model with modal decomposition outperformed the others in prediction accuracy.For soybean and maize NDVI prediction, the model achieved MAPE of 2.111% and 2.425%, RMSE of 0.012, and R 2 of 0.971 and 0.968, respectively.This model provides a robust foundation for real-time management and precision fertilization, ensuring optimal crop growth and yield.