Modeling Agricultural Problems Based on Neural Technologies

Models for predicting soil moisture were built based on agronomic and climatic data. Transfer learning models such as DenseNet, ResNet, and EfficientNet were used to identify complex nonlinear relationships between data. Improved model performance was achieved through regularization methods and tuning of model training hyperparameters. Smoother model learning curves were obtained, demonstrating the improved generalization ability of regularized models. To optimize the models, interactions between data features were analyzed. Correlations between features were constructed for model construction. Optimized models were constructed that take into account complex relationships in the data to build effective soil moisture forecasting models. A strong correlation was found between soil type and fertilizers. Climate factors such as temperature and air humidity were shown to interact nonlinearly. Nutrients such as N-nitrogen, P-phosphorus, and K-potassium exhibit complex interactions. The hybrid neural network models constructed with interaction blocks based on transfer learning and dense layers (Dense, ResNet, and EfficientNet) demonstrated the best results. Using a comprehensive analysis, neural network-based models were built with high performance compared to classical machine learning methods.