Dynamic Prediction Model of Crop Canopy Temperature Based on VMD-LSTM

[Objective] Accurate prediction of crop canopy temperature is essential for comprehensively assessing crop growth status and guiding agricultural production. This study focuses on kiwifruit and grapes to address the challenges in accurately predicting crop canopy temperature. [Methods] A dynamic prediction model for crop canopy temperature was developed based on Long Short-Term Memory (LSTM), Variational Mode Decomposition (VMD), and the Rime Ice Morphology-based Optimization Algorithm (RIME) optimization algorithm, named RIME-VMD-RIME-LSTM (RIME2-VMD-LSTM). Firstly, crop canopy temperature data were collected by an inspection robot suspended on a cableway. Secondly, through the performance of multiple pre-test experiments, VMD-LSTM was selected as the base model. To reduce cross-interference between different frequency components of VMD, the K-means clustering algorithm was applied to cluster the sample entropy of each component, reconstructing them into new components. Finally, the RIME optimization algorithm was utilized to optimize the parameters of VMD and LSTM, enhancing the model's prediction accuracy. [Results and Discussions] The experimental results demonstrated that the proposed model achieved lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) (0.360 1 and 0.254 3 °C, respectively) in modeling different noise environments than the comparator model. Furthermore, the R2 value reached a maximum of 0.994 7. [Conclusions] This model provides a feasible method for dynamically predicting crop canopy temperature and offers data support for assessing crop growth status in agricultural parks.