Dynamic parameter calibration based deep network for paddy yield prediction

Abstract

Paddy yield prediction plays a crucial role in agriculture, enabling farmers to make informed decisions. This work proposes an innovative approach combining Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) for accurate paddy yield forecasting. The hybrid model harnesses the spatial understanding capabilities of CNNs and the sequential learning ability of RNNs to capture both local and temporal dependencies in agricultural data. A key enhancement introduced in this method is the incorporation of a dynamic parameter calibration technique. Traditional regularization methods often rely on static values, which may not adapt effectively to varying complexities in the dataset. The proposed approach dynamically adjusts the regularization strength parameters during training, allowing the model to better converge to different patterns and fluctuations in paddy growth parameters. The dataset utilized for training and evaluation comprises comprehensive agricultural variables, including soil composition, climate conditions, and historical yield data. Experiments demonstrate the effectiveness of the hybrid CNN-RNN architecture with dynamic parameter calibration which improves prediction accuracy over conventional models. This research contributes to the advancement of precision agriculture by providing a more robust and adaptable framework for paddy yield prediction. The integration of spatial and temporal features, along with dynamic parameter calibration, showcases the potential for optimizing agricultural decision-making processes and mitigating the impact of unpredictable factors on paddy production.